Repository navigation
Tags
Tags
Version Packages (canary) (vercel#16024) This PR was opened by the [Changesets release](https://github.com/changesets/action) GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to main, this PR will be updated.⚠️ ⚠️ ⚠️ ⚠️ ⚠️ ⚠️ `main` is currently in **pre mode** so this branch has prereleases rather than normal releases. If you want to exit prereleases, run `changeset pre exit` on `main`.⚠️ ⚠️ ⚠️ ⚠️ ⚠️ ⚠️ # Releases ## ai@7.0.0-canary.172 ### Patch Changes - 25a64f8: Remove deprecated experimental generateImage exports. - 375fdd7: fix: harden download URL SSRF guard against hostname and redirect bypasses `validateDownloadUrl` and the file download helpers (`downloadBlob`, `download`) could be bypassed in several ways when handling untrusted URLs: - A fully-qualified hostname with a trailing dot (e.g. `localhost.`, `myhost.local.`) skipped the localhost/`.local` blocklist. - IPv6 addresses that embed an IPv4 address in their last 32 bits — IPv4-compatible (`::127.0.0.1`), IPv4-translated (`::ffff:0:127.0.0.1`), and NAT64 (`64:ff9b::127.0.0.1`, including the `64:ff9b:1::/48` local-use prefix) — were not decoded and checked against the private IPv4 ranges. - Redirects were validated only _after_ `fetch` had already followed them, so the request to a redirect target (e.g. an internal/metadata address) had already been issued before the check ran. - Several reserved/internal address ranges were not blocked: CGNAT (`100.64.0.0/10`, used by some cloud providers for internal traffic), benchmarking (`198.18.0.0/15`), IETF protocol assignments (`192.0.0.0/24`), the reserved `240.0.0.0/4` block (including the `255.255.255.255` broadcast address), and IPv6 site-local (`fec0::/10`) and multicast (`ff00::/8`). The validator now strips trailing dots before the hostname checks and fully expands IPv6 addresses to detect embedded private IPv4 targets. The download helpers now follow redirects manually (`redirect: 'manual'`), re-validating each hop before requesting it, so an unsafe redirect target is never fetched. When a redirect cannot be inspected because the runtime returns an opaque response, the helpers fail closed (reject the redirect) on the server; only in a real browser — where SSRF is not reachable (fetch is constrained by CORS and cannot reach a server's internal network or cloud-metadata endpoints) — is the redirect followed natively so legitimate redirected downloads keep working. - f18b08f: fix: redact server error details from UI message streams by default `toUIMessageStream`, `createUIMessageStream`, and `toUIMessageChunk` defaulted their `onError` callback to `getErrorMessage`, which serializes the raw error (`error.toString()` / `JSON.stringify(error)`) into the client-facing `{ type: 'error', errorText }` chunk — and also into `tool-output-error` parts. The documented default was `() => 'An error occurred.'`, so applications relying on the documented behavior were unknowingly streaming server exception details (internal hostnames, paths, provider request data, validation inputs) to end users. The default `onError` now returns the documented generic `'An error occurred.'`. Raw error details are only emitted when the developer explicitly supplies an `onError` handler. This also redacts `tool-output-error` and invalid-tool-input error text by default; pass an `onError` to surface richer messages. - b4507d5: fix(provider-utils): cancel response body on download rejection to prevent socket leak When a download was rejected early — because the `Content-Length` header exceeded the size limit, the response status was not ok, or a redirect resolved to a blocked URL — the fetch response body was left unconsumed and uncancelled. With WHATWG Fetch/undici this leaves the underlying TCP socket open instead of returning it to the connection pool, allowing an attacker-controlled origin to exhaust file descriptors and cause a denial of service. The body is now cancelled on all early-rejection paths in `readResponseWithSizeLimit`, `download`, and `downloadBlob`, and `fetchWithValidatedRedirects` cancels each redirect hop's body before following or rejecting the next hop. - Updated dependencies [8c17bf8] - Updated dependencies [aeda373] - Updated dependencies [558777f] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/gateway@4.0.0-canary.104 - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/alibaba@2.0.0-canary.59 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/amazon-bedrock@5.0.0-canary.84 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/anthropic@4.0.0-canary.66 - @ai-sdk/openai@4.0.0-canary.73 ## @ai-sdk/angular@3.0.0-canary.172 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/anthropic@4.0.0-canary.66 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/anthropic-aws@1.0.0-canary.7 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/anthropic@4.0.0-canary.66 ## @ai-sdk/assemblyai@3.0.0-canary.50 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/azure@4.0.0-canary.75 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/deepseek@3.0.0-canary.54 - @ai-sdk/openai@4.0.0-canary.73 ## @ai-sdk/baseten@2.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/black-forest-labs@2.0.0-canary.50 ### Patch Changes - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/bytedance@2.0.0-canary.51 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/cerebras@3.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/cohere@4.0.0-canary.52 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/deepgram@3.0.0-canary.49 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/deepinfra@3.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/deepseek@3.0.0-canary.54 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/devtools@1.0.0-canary.30 ### Patch Changes - 32ce45c: fix(devtools): secure devtools viewer api access - 275674e: fix(devtools): validate viewer database paths ## @ai-sdk/elevenlabs@3.0.0-canary.49 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/fal@3.0.0-canary.49 ### Patch Changes - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/fireworks@3.0.0-canary.56 ### Patch Changes - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/gateway@4.0.0-canary.104 ### Patch Changes - 8c17bf8: fix(gateway): surface provider warnings in embedding and reranking responses - 558777f: fix(gateway): accept deprecated warnings in image, speech, transcription, and video responses - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/gladia@3.0.0-canary.49 ### Patch Changes - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/google@4.0.0-canary.81 ### Patch Changes - 5878b40: fix(google): prevent prototype pollution when streaming tool args - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/google-vertex@5.0.0-canary.107 ### Patch Changes - Updated dependencies [5878b40] - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/google@4.0.0-canary.81 - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/anthropic@4.0.0-canary.66 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/groq@4.0.0-canary.53 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/harness@1.0.0-canary.7 ### Patch Changes - 3d87086: fix(harness): guard against invalid resuming a session vs continuing a turn - 1ea15a3: fix(harness): fix various bugs with harness skills not being correctly processed by the harness adapters - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/harness-claude-code@1.0.0-canary.3 ### Patch Changes - 1ea15a3: fix(harness): fix various bugs with harness skills not being correctly processed by the harness adapters - e551763: fix(harness): avoid using peer dependencies for underlying harness and sandbox SDKs - Updated dependencies [3d87086] - Updated dependencies [aeda373] - Updated dependencies [1ea15a3] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/harness@1.0.0-canary.7 - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/harness-codex@1.0.0-canary.3 ### Patch Changes - 1ea15a3: fix(harness): fix various bugs with harness skills not being correctly processed by the harness adapters - e551763: fix(harness): avoid using peer dependencies for underlying harness and sandbox SDKs - Updated dependencies [3d87086] - Updated dependencies [aeda373] - Updated dependencies [1ea15a3] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/harness@1.0.0-canary.7 - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/harness-pi@1.0.0-canary.3 ### Patch Changes - 1ea15a3: fix(harness): fix various bugs with harness skills not being correctly processed by the harness adapters - e551763: fix(harness): avoid using peer dependencies for underlying harness and sandbox SDKs - Updated dependencies [3d87086] - Updated dependencies [aeda373] - Updated dependencies [1ea15a3] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/harness@1.0.0-canary.7 - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/huggingface@2.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/hume@3.0.0-canary.49 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/klingai@4.0.0-canary.50 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/langchain@3.0.0-canary.172 ### Patch Changes - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - ai@7.0.0-canary.172 ## @ai-sdk/llamaindex@3.0.0-canary.172 ### Patch Changes - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - ai@7.0.0-canary.172 ## @ai-sdk/lmnt@3.0.0-canary.49 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/luma@3.0.0-canary.49 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/mcp@2.0.0-canary.63 ### Patch Changes - 2a150f8: fix(mcp): lock first sse endpoint received via event - b44b051: fix(mcp): prevent prototype-named tools from bypassing the `schemas` allowlist When using `client.tools({ schemas })` to expose only an explicitly allowed subset of an MCP server's tools, the allowlist check used the `in` operator, which also matches inherited `Object.prototype` properties. A server-advertised tool named `constructor`, `toString`, `__proto__`, etc. would pass the check even though the developer never defined it in `schemas`, and was then exposed to the model and executable. The check now uses `Object.hasOwn`, so only explicitly defined tools are returned. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/mistral@4.0.0-canary.54 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/moonshotai@3.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/open-responses@2.0.0-canary.53 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/openai@4.0.0-canary.73 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/openai-compatible@3.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/otel@1.0.0-canary.118 ### Patch Changes - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - ai@7.0.0-canary.172 ## @ai-sdk/perplexity@4.0.0-canary.52 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/policy-opa@1.0.0-canary.9 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/prodia@2.0.0-canary.52 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/provider-utils@5.0.0-canary.48 ### Patch Changes - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - 375fdd7: fix: harden download URL SSRF guard against hostname and redirect bypasses `validateDownloadUrl` and the file download helpers (`downloadBlob`, `download`) could be bypassed in several ways when handling untrusted URLs: - A fully-qualified hostname with a trailing dot (e.g. `localhost.`, `myhost.local.`) skipped the localhost/`.local` blocklist. - IPv6 addresses that embed an IPv4 address in their last 32 bits — IPv4-compatible (`::127.0.0.1`), IPv4-translated (`::ffff:0:127.0.0.1`), and NAT64 (`64:ff9b::127.0.0.1`, including the `64:ff9b:1::/48` local-use prefix) — were not decoded and checked against the private IPv4 ranges. - Redirects were validated only _after_ `fetch` had already followed them, so the request to a redirect target (e.g. an internal/metadata address) had already been issued before the check ran. - Several reserved/internal address ranges were not blocked: CGNAT (`100.64.0.0/10`, used by some cloud providers for internal traffic), benchmarking (`198.18.0.0/15`), IETF protocol assignments (`192.0.0.0/24`), the reserved `240.0.0.0/4` block (including the `255.255.255.255` broadcast address), and IPv6 site-local (`fec0::/10`) and multicast (`ff00::/8`). The validator now strips trailing dots before the hostname checks and fully expands IPv6 addresses to detect embedded private IPv4 targets. The download helpers now follow redirects manually (`redirect: 'manual'`), re-validating each hop before requesting it, so an unsafe redirect target is never fetched. When a redirect cannot be inspected because the runtime returns an opaque response, the helpers fail closed (reject the redirect) on the server; only in a real browser — where SSRF is not reachable (fetch is constrained by CORS and cannot reach a server's internal network or cloud-metadata endpoints) — is the redirect followed natively so legitimate redirected downloads keep working. - b4507d5: fix(provider-utils): cancel response body on download rejection to prevent socket leak When a download was rejected early — because the `Content-Length` header exceeded the size limit, the response status was not ok, or a redirect resolved to a blocked URL — the fetch response body was left unconsumed and uncancelled. With WHATWG Fetch/undici this leaves the underlying TCP socket open instead of returning it to the connection pool, allowing an attacker-controlled origin to exhaust file descriptors and cause a denial of service. The body is now cancelled on all early-rejection paths in `readResponseWithSizeLimit`, `download`, and `downloadBlob`, and `fetchWithValidatedRedirects` cancels each redirect hop's body before following or rejecting the next hop. ## @ai-sdk/quiverai@2.0.0-canary.5 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/react@4.0.0-canary.175 ### Patch Changes - 555c5de: fix(react): deny MCP App tool calls by default when allowedTools is omitted `experimental_MCPAppRenderer`'s bridge only enforced the `allowedTools` allowlist when it was non-null, so omitting `allowedTools` skipped the check and forwarded every `tools/call` from the (untrusted) MCP App iframe to the host's `callTool`. A malicious or compromised MCP server could therefore invoke any tool the host wired up. Tool invocation from MCP App content is now deny-by-default: if `allowedTools` is not explicitly provided, all `tools/call` requests are rejected. To expose tools to an app, list them in `handlers.allowedTools`. - Updated dependencies [2a150f8] - Updated dependencies [b44b051] - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/mcp@2.0.0-canary.63 - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/replicate@3.0.0-canary.50 ### Patch Changes - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/revai@3.0.0-canary.50 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/rsc@3.0.0-canary.173 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/sandbox-just-bash@1.0.0-canary.7 ### Patch Changes - e551763: fix(harness): avoid using peer dependencies for underlying harness and sandbox SDKs - Updated dependencies [3d87086] - Updated dependencies [aeda373] - Updated dependencies [1ea15a3] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/harness@1.0.0-canary.7 - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/sandbox-vercel@1.0.0-canary.7 ### Patch Changes - e551763: fix(harness): avoid using peer dependencies for underlying harness and s 8000 andbox SDKs - Updated dependencies [3d87086] - Updated dependencies [aeda373] - Updated dependencies [1ea15a3] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/harness@1.0.0-canary.7 - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/svelte@5.0.0-canary.172 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/togetherai@3.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/tui@1.0.0-canary.7 ### Patch Changes - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - ai@7.0.0-canary.172 ## @ai-sdk/valibot@3.0.0-canary.48 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/vercel@3.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/voyage@2.0.0-canary.23 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/vue@4.0.0-canary.172 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/workflow@1.0.0-canary.89 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/xai@4.0.0-canary.74 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Version Packages (vercel#16026) This PR was opened by the [Changesets release](https://github.com/changesets/action) GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to release-v6.0, this PR will be updated. # Releases ## ai@6.0.203 ### Patch Changes - f42aa79: fix: harden download URL SSRF guard against hostname and redirect bypasses `validateDownloadUrl` and the file download helpers (`downloadBlob`, `download`) could be bypassed in several ways when handling untrusted URLs: - A fully-qualified hostname with a trailing dot (e.g. `localhost.`, `myhost.local.`) skipped the localhost/`.local` blocklist. - IPv6 addresses that embed an IPv4 address in their last 32 bits — IPv4-compatible (`::127.0.0.1`), IPv4-translated (`::ffff:0:127.0.0.1`), and NAT64 (`64:ff9b::127.0.0.1`, including the `64:ff9b:1::/48` local-use prefix) — were not decoded and checked against the private IPv4 ranges. - Redirects were validated only _after_ `fetch` had already followed them, so the request to a redirect target (e.g. an internal/metadata address) had already been issued before the check ran. - Several reserved/internal address ranges were not blocked: CGNAT (`100.64.0.0/10`, used by some cloud providers for internal traffic), benchmarking (`198.18.0.0/15`), IETF protocol assignments (`192.0.0.0/24`), the reserved `240.0.0.0/4` block (including the `255.255.255.255` broadcast address), and IPv6 site-local (`fec0::/10`) and multicast (`ff00::/8`). The validator now strips trailing dots before the hostname checks and fully expands IPv6 addresses to detect embedded private IPv4 targets. The download helpers now follow redirects manually (`redirect: 'manual'`), re-validating each hop before requesting it, so an unsafe redirect target is never fetched. When a redirect cannot be inspected because the runtime returns an opaque response, the helpers fail closed (reject the redirect) on the server; only in a real browser — where SSRF is not reachable (fetch is constrained by CORS and cannot reach a server's internal network or cloud-metadata endpoints) — is the redirect followed natively so legitimate redirected downloads keep working. - 5291f7e: Harden stream text processing and middleware against prototype pollution from stream part IDs. - b4b575a: fix: redact server error details from UI message streams by default `streamText(...).toUIMessageStream()` and `createUIMessageStream` defaulted their `onError` callback to `getErrorMessage`, which serializes the raw error (`error.toString()` / `JSON.stringify(error)`) into the client-facing `{ type: 'error', errorText }` chunk — and also into `tool-output-error` parts. The documented default was `() => 'An error occurred.'`, so applications relying on the documented behavior were unknowingly streaming server exception details (internal hostnames, paths, provider request data, validation inputs) to end users. The default `onError` now returns the documented generic `'An error occurred.'`. Raw error details are only emitted when the developer explicitly supplies an `onError` handler. This also redacts `tool-output-error` and invalid-tool-input error text by default; pass an `onError` to surface richer messages. - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/gateway@3.0.129 ## @ai-sdk/alibaba@1.0.28 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/openai-compatible@2.0.50 ## @ai-sdk/amazon-bedrock@4.0.117 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/anthropic@3.0.84 - @ai-sdk/openai@3.0.71 ## @ai-sdk/angular@2.0.204 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - Updated dependencies [5291f7e] - Updated dependencies [b4b575a] - @ai-sdk/provider-utils@4.0.29 - ai@6.0.203 ## @ai-sdk/anthropic@3.0.84 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/anthropic-aws@1.0.6 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/anthropic@3.0.84 ## @ai-sdk/assemblyai@2.0.35 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/azure@3.0.74 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/deepseek@2.0.38 - @ai-sdk/openai@3.0.71 ## @ai-sdk/baseten@1.0.53 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/openai-compatible@2.0.50 ## @ai-sdk/black-forest-labs@1.0.36 ### Patch Changes - bfa5864: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/bytedance@1.0.17 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/cerebras@2.0.56 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/openai-compatible@2.0.50 ## @ai-sdk/cohere@3.0.38 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/deepgram@2.0.35 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/deepinfra@2.0.54 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/openai-compatible@2.0.50 ## @ai-sdk/deepseek@2.0.38 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/devtools@0.0.19 ### Patch Changes - f27afa8: fix(devtools): secure devtools viewer api access ## @ai-sdk/elevenlabs@2.0.35 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/fal@2.0.36 ### Patch Changes - bfa5864: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/fireworks@2.0.55 ### Patch Changes - bfa5864: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/openai-compatible@2.0.50 ## @ai-sdk/gateway@3.0.129 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/gladia@2.0.35 ### Patch Changes - bfa5864: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/google@3.0.82 ### Patch Changes - 3258f22: fix(google): prevent prototype pollution when streaming tool args - bfa5864: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/google-vertex@4.0.145 ### Patch Changes - Updated dependencies [3258f22] - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/google@3.0.82 - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/anthropic@3.0.84 - @ai-sdk/openai-compatible@2.0.50 ## @ai-sdk/groq@3.0.41 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/huggingface@1.0.52 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/openai-compatible@2.0.50 ## @ai-sdk/hume@2.0.35 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/klingai@3.0.20 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/langchain@2.0.210 ### Patch Changes - Updated dependencies [f42aa79] - Updated dependencies [5291f7e] - Updated dependencies [b4b575a] - ai@6.0.203 ## @ai-sdk/llamaindex@2.0.203 ### Patch Changes - Updated dependencies [f42aa79] - Updated dependencies [5291f7e] - Updated dependencies [b4b575a] - ai@6.0.203 ## @ai-sdk/lmnt@2.0.35 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/luma@2.0.35 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/mcp@1.0.49 ### Patch Changes - 3e8d9ba: fix(mcp): lock first sse endpoint received via event - 4fa7354: fix(mcp): prevent prototype-named tools from bypassing the `schemas` allowlist When using `client.tools({ schemas })` to expose only an explicitly allowed subset of an MCP server's tools, the allowlist check used the `in` operator, which also matches inherited `Object.prototype` properties. A server-advertised tool named `constructor`, `toString`, `__proto__`, etc. would pass the check even though the developer never defined it in `schemas`, and was then exposed to the model and executable. The check now uses `Object.hasOwn`, so only explicitly defined tools are returned. - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/mistral@3.0.39 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/moonshotai@2.0.25 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/openai-compatible@2.0.50 ## @ai-sdk/open-responses@1.0.18 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/openai@3.0.71 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/openai-compatible@2.0.50 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/perplexity@3.0.35 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/prodia@1.0.34 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/provider-utils@4.0.29 ### Patch Changes - bfa5864: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - f42aa79: fix: harden download URL SSRF guard against hostname and redirect bypasses `validateDownloadUrl` and the file download helpers (`downloadBlob`, `download`) could be bypassed in several ways when handling untrusted URLs: - A fully-qualified hostname with a trailing dot (e.g. `localhost.`, `myhost.local.`) skipped the localhost/`.local` blocklist. - IPv6 addresses that embed an IPv4 address in their last 32 bits — IPv4-compatible (`::127.0.0.1`), IPv4-translated (`::ffff:0:127.0.0.1`), and NAT64 (`64:ff9b::127.0.0.1`, including the `64:ff9b:1::/48` local-use prefix) — were not decoded and checked against the private IPv4 ranges. - Redirects were validated only _after_ `fetch` had already followed them, so the request to a redirect target (e.g. an internal/metadata address) had already been issued before the check ran. - Several reserved/internal address ranges were not blocked: CGNAT (`100.64.0.0/10`, used by some cloud providers for internal traffic), benchmarking (`198.18.0.0/15`), IETF protocol assignments (`192.0.0.0/24`), the reserv 8000 ed `240.0.0.0/4` block (including the `255.255.255.255` broadcast address), and IPv6 site-local (`fec0::/10`) and multicast (`ff00::/8`). The validator now strips trailing dots before the hostname checks and fully expands IPv6 addresses to detect embedded private IPv4 targets. The download helpers now follow redirects manually (`redirect: 'manual'`), re-validating each hop before requesting it, so an unsafe redirect target is never fetched. When a redirect cannot be inspected because the runtime returns an opaque response, the helpers fail closed (reject the redirect) on the server; only in a real browser — where SSRF is not reachable (fetch is constrained by CORS and cannot reach a server's internal network or cloud-metadata endpoints) — is the redirect followed natively so legitimate redirected downloads keep working. ## @ai-sdk/quiverai@1.0.2 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/react@3.0.205 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - Updated dependencies [5291f7e] - Updated dependencies [b4b575a] - @ai-sdk/provider-utils@4.0.29 - ai@6.0.203 ## @ai-sdk/replicate@2.0.35 ### Patch Changes - bfa5864: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/revai@2.0.35 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/rsc@2.0.203 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - Updated dependencies [5291f7e] - Updated dependencies [b4b575a] - @ai-sdk/provider-utils@4.0.29 - ai@6.0.203 ## @ai-sdk/svelte@4.0.203 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - Updated dependencies [5291f7e] - Updated dependencies [b4b575a] - @ai-sdk/provider-utils@4.0.29 - ai@6.0.203 ## @ai-sdk/togetherai@2.0.55 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/openai-compatible@2.0.50 ## @ai-sdk/valibot@2.0.30 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/vercel@2.0.52 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/openai-compatible@2.0.50 ## @ai-sdk/voyage@1.0.6 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/vue@3.0.203 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - Updated dependencies [5291f7e] - Updated dependencies [b4b575a] - @ai-sdk/provider-utils@4.0.29 - ai@6.0.203 ## @ai-sdk/xai@3.0.95 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/openai-compatible@2.0.50 Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Version Packages (vercel#16027) This PR was opened by the [Changesets release](https://github.com/changesets/action) GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to release-v5.0, this PR will be updated. # Releases ## ai@5.0.200 ### Patch Changes - eea9166: fix: harden download URL SSRF guard against hostname and redirect bypasses `validateDownloadUrl` and the file download helpers (`downloadBlob`, `download`) could be bypassed in several ways when handling untrusted URLs: - A fully-qualified hostname with a trailing dot (e.g. `localhost.`, `myhost.local.`) skipped the localhost/`.local` blocklist. - IPv6 addresses that embed an IPv4 address in their last 32 bits — IPv4-compatible (`::127.0.0.1`), IPv4-translated (`::ffff:0:127.0.0.1`), and NAT64 (`64:ff9b::127.0.0.1`, including the `64:ff9b:1::/48` local-use prefix) — were not decoded and checked against the private IPv4 ranges. - Redirects were validated only _after_ `fetch` had already followed them, so the request to a redirect target (e.g. an internal/metadata address) had already been issued before the check ran. - Several reserved/internal address ranges were not blocked: CGNAT (`100.64.0.0/10`, used by some cloud providers for internal traffic), benchmarking (`198.18.0.0/15`), IETF protocol assignments (`192.0.0.0/24`), the reserved `240.0.0.0/4` block (including the `255.255.255.255` broadcast address), and IPv6 site-local (`fec0::/10`) and multicast (`ff00::/8`). The validator now strips trailing dots before the hostname checks and fully expands IPv6 addresses to detect embedded private IPv4 targets. The download helpers now follow redirects manually (`redirect: 'manual'`), re-validating each hop before requesting it, so an unsafe redirect target is never fetched. When a redirect cannot be inspected because the runtime returns an opaque response, the helpers fail closed (reject the redirect) on the server; only in a real browser — where SSRF is not reachable (fetch is constrained by CORS and cannot reach a server's internal network or cloud-metadata endpoints) — is the redirect followed natively so legitimate redirected downloads keep working. - 15cf74f: Harden stream text processing and middleware against prototype pollution from stream part IDs. - 8ad6f80: fix: redact server error details from UI message streams by default `streamText(...).toUIMessageStream()` and `createUIMessageStream` defaulted their `onError` callback to `getErrorMessage`, which serializes the raw error (`error.toString()` / `JSON.stringify(error)`) into the client-facing `{ type: 'error', errorText }` chunk — and also into `tool-output-error` parts. The documented default was `() => 'An error occurred.'`, so applications relying on the documented behavior were unknowingly streaming server exception details (internal hostnames, paths, provider request data, validation inputs) to end users. The default `onError` now returns the documented generic `'An error occurred.'`. Raw error details are only emitted when the developer explicitly supplies an `onError` handler. This also redacts `tool-output-error` and invalid-tool-input error text by default; pass an `onError` to surface richer messages. - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/gateway@2.0.99 ## @ai-sdk/alibaba@0.0.12 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/openai-compatible@1.0.40 ## @ai-sdk/amazon-bedrock@3.0.102 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/anthropic@2.0.82 ## @ai-sdk/angular@1.0.200 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - Updated dependencies [15cf74f] - Updated dependencies [8ad6f80] - @ai-sdk/provider-utils@3.0.26 - ai@5.0.200 ## @ai-sdk/anthropic@2.0.82 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/assemblyai@1.0.30 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/azure@2.0.111 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/deepseek@1.0.43 - @ai-sdk/openai@2.0.107 ## @ai-sdk/baseten@0.0.23 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/openai-compatible@1.0.40 ## @ai-sdk/black-forest-labs@0.0.15 ### Patch Changes - 9f67efe: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/cerebras@1.0.45 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/openai-compatible@1.0.40 ## @ai-sdk/cohere@2.0.30 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ 8000 ai-sdk/provider-utils@3.0.26 ## @ai-sdk/deepgram@1.0.29 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/deepinfra@1.0.43 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/openai-compatible@1.0.40 ## @ai-sdk/deepseek@1.0.43 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/elevenlabs@1.0.29 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/fal@1.0.33 ### Patch Changes - 9f67efe: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/fireworks@1.0.41 ### Patch Changes - 9f67efe: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/openai-compatible@1.0.40 ## @ai-sdk/gateway@2.0.99 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/gladia@1.0.28 ### Patch Changes - 9f67efe: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/google@2.0.75 ### Patch Changes - 9f67efe: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/google-vertex@3.0.142 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/google@2.0.75 - @ai-sdk/anthropic@2.0.82 - @ai-sdk/openai-compatible@1.0.40 ## @ai-sdk/groq@2.0.41 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/huggingface@0.0.23 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/openai-compatible@1.0.40 ## @ai-sdk/hume@1.0.28 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/langchain@1.0.200 ### Patch Changes - Updated dependencies [eea9166] - Updated dependencies [15cf74f] - Updated dependencies [8ad6f80] - ai@5.0.200 ## @ai-sdk/llamaindex@1.0.200 ### Patch Changes - Updated dependencies [eea9166] - Updated dependencies [15cf74f] - Updated dependencies [8ad6f80] - ai@5.0.200 ## @ai-sdk/lmnt@1.0.28 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/luma@1.0.28 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/mcp@0.0.19 ### Patch Changes - d872a7a: fix(mcp): lock first sse endpoint received via event - f4cd468: fix(mcp): prevent prototype-named tools from bypassing the `schemas` allowlist When using `client.tools({ schemas })` to expose only an explicitly allowed subset of an MCP server's tools, the allowlist check used the `in` operator, which also matches inherited `Object.prototype` properties. A server-advertised tool named `constructor`, `toString`, `__proto__`, etc. would pass the check even though the developer never defined it in `schemas`, and was then exposed to the model and executable. The check now uses `Object.hasOwn`, so only explicitly defined tools are returned. - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/mistral@2.0.34 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/moonshotai@0.0.11 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/openai-compatible@1.0.40 ## @ai-sdk/openai@2.0.107 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/openai-compatible@1.0.40 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/perplexity@2.0.31 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/prodia@0.0.10 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/provider-utils@3.0.26 ### Patch Changes - 9f67efe: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - eea9166: fix: harden download URL SSRF guard against hostname and redirect bypasses `validateDownloadUrl` and the file download helpers (`downloadBlob`, `download`) could be bypassed in several ways when handling untrusted URLs: - A fully-qualified hostname with a trailing dot (e.g. `localhost.`, `myhost.local.`) skipped the localhost/`.local` blocklist. - IPv6 addresses that embed an IPv4 address in their last 32 bits — IPv4-compatible (`::127.0.0.1`), IPv4-translated (`::ffff:0:127.0.0.1`), and NAT64 (`64:ff9b::127.0.0.1`, including the `64:ff9b:1::/48` local-use prefix) — were not decoded and checked against the private IPv4 ranges. - Redirects were validated only _after_ `fetch` had already followed them, so the request to a redirect target (e.g. an internal/metadata address) had already been issued before the check ran. - Several reserved/internal address ranges were not blocked: CGNAT (`100.64.0.0/10`, used by some cloud providers for internal traffic), benchmarking (`198.18.0.0/15`), IETF protocol assignments (`192.0.0.0/24`), the reserved `240.0.0.0/4` block (including the `255.255.255.255` broadcast address), and IPv6 site-local (`fec0::/10`) and multicast (`ff00::/8`). The validator now strips trailing dots before the hostname checks and fully expands IPv6 addresses to detect embedded private IPv4 targets. The download helpers now follow redirects manually (`redirect: 'manual'`), re-validating each hop before requesting it, so an unsafe redirect target is never fetched. When a redirect cannot be inspected because the runtime returns an opaque response, the helpers fail closed (reject the redirect) on the server; only in a real browser — where SSRF is not reachable (fetch is constrained by CORS and cannot reach a server's internal network or cloud-metadata endpoints) — is the redirect followed natively so legitimate redirected downloads keep working. ## @ai-sdk/react@2.0.202 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - Updated dependencies [15cf74f] - Updated dependencies [8ad6f80] - @ai-sdk/provider-utils@3.0.26 - ai@5.0.200 ## @ai-sdk/replicate@1.0.28 ### Patch Changes - 9f67efe: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/revai@1.0.28 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/rsc@1.0.202 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - Updated dependencies [15cf74f] - Updated dependencies [8ad6f80] - @ai-sdk/provider-utils@3.0.26 - ai@5.0.200 ## @ai-sdk/svelte@3.0.200 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - Updated dependencies [15cf74f] - Updated dependencies [8ad6f80] - @ai-sdk/provider-utils@3.0.26 - ai@5.0.200 ## @ai-sdk/togetherai@1.0.43 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/openai-compatible@1.0.40 ## @ai-sdk/valibot@1.0.26 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/vercel@1.0.41 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/openai-compatible@1.0.40 ## @ai-sdk/vue@2.0.200 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - Updated dependencies [15cf74f] - Updated dependencies [8ad6f80] - @ai-sdk/provider-utils@3.0.26 - ai@5.0.200 ## @ai-sdk/xai@2.0.74 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/openai-compatible@1.0.40 Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Version Packages (canary) (vercel#16084) This PR was opened by the [Changesets release](https://github.com/changesets/action) GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to main, this PR will be updated.⚠️ ⚠️ ⚠️ ⚠️ ⚠️ ⚠️ `main` is currently in **pre mode** so this branch has prereleases rather than normal releases. If you want to exit prereleases, run `changeset pre exit` on `main`.⚠️ ⚠️ ⚠️ ⚠️ ⚠️ ⚠️ # Releases ## ai@7.0.0-canary.173 ### Patch Changes - Updated dependencies [efec111] - @ai-sdk/gateway@4.0.0-canary.105 ## @ai-sdk/angular@3.0.0-canary.173 ### Patch Changes - ai@7.0.0-canary.173 ## @ai-sdk/fireworks@3.0.0-canary.57 ### Patch Changes - 38010a1: Enable `includeUsage` for Fireworks so streaming responses report token usage ## @ai-sdk/gateway@4.0.0-canary.105 ### Patch Changes - efec111: chore(provider/gateway): update gateway model settings files ## @ai-sdk/harness@1.0.0-canary.9 ### Patch Changes - ai@7.0.0-canary.173 ## @ai-sdk/harness-claude-code@1.0.0-canary.5 ### Patch Changes - @ai-sdk/harness@1.0.0-canary.9 ## @ai-sdk/harness-codex@1.0.0-canary.5 ### Patch Changes - @ai-sdk/harness@1.0.0-canary.9 ## @ai-sdk/harness-pi@1.0.0-canary.5 ### Patch Changes - @ai-sdk/harness@1.0.0-canary.9 ## @ai-sdk/langchain@3.0.0-canary.173 ### Patch Changes - c1afaed: fix(langchain): prevent polluting global object.prototype - ai@7.0.0-canary.173 ## @ai-sdk/llamaindex@3.0.0-canary.173 ### Patch Changes - ai@7.0.0-canary.173 ## @ai-sdk/mcp@2.0.0-canary.64 ### Patch Changes - 024a6b4: fix(mcp): validate oauth metadata issuer during discovery ## @ai-sdk/otel@1.0.0-canary.119 ### Patch Changes - ai@7.0.0-canary.173 ## @ai-sdk/policy-opa@1.0.0-canary.10 ### Patch Changes - ai@7.0.0-canary.173 ## @ai-sdk/react@4.0.0-canary.176 ### Patch Changes - Updated dependencies [024a6b4] - @ai-sdk/mcp@2.0.0-canary.64 - ai@7.0.0-canary.173 ## @ai-sdk/rsc@3.0.0-canary.174 ### Patch Changes - ai@7.0.0-canary.173 ## @ai-sdk/sandbox-just-bash@1.0.0-canary.9 ### Patch Changes - @ai-sdk/harness@1.0.0-canary.9 ## @ai-sdk/sandbox-vercel@1.0.0-canary.9 ### Patch Changes - @ai-sdk/harness@1.0.0-canary.9 ## @ai-sdk/svelte@5.0.0-canary.173 ### Patch Changes - ai@7.0.0-canary.173 ## @ai-sdk/tui@1.0.0-canary.8 ### Patch Changes - ai@7.0.0-canary.173 ## @ai-sdk/vue@4.0.0-canary.173 ### Patch Changes - ai@7.0.0-canary.173 ## @ai-sdk/workflow@1.0.0-canary.90 ### Patch Changes - ai@7.0.0-canary.173 Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Version Packages (canary) (vercel#16024) This PR was opened by the [Changesets release](https://github.com/changesets/action) GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to main, this PR will be updated.⚠️ ⚠️ ⚠️ ⚠️ ⚠️ ⚠️ `main` is currently in **pre mode** so this branch has prereleases rather than normal releases. If you want to exit prereleases, run `changeset pre exit` on `main`.⚠️ ⚠️ ⚠️ ⚠️ ⚠️ ⚠️ # Releases ## ai@7.0.0-canary.172 ### Patch Changes - 25a64f8: Remove deprecated experimental generateImage exports. - 375fdd7: fix: harden download URL SSRF guard against hostname and redirect bypasses `validateDownloadUrl` and the file download helpers (`downloadBlob`, `download`) could be bypassed in several ways when handling untrusted URLs: - A fully-qualified hostname with a trailing dot (e.g. `localhost.`, `myhost.local.`) skipped the localhost/`.local` blocklist. - IPv6 addresses that embed an IPv4 address in their last 32 bits — IPv4-compatible (`::127.0.0.1`), IPv4-translated (`::ffff:0:127.0.0.1`), and NAT64 (`64:ff9b::127.0.0.1`, including the `64:ff9b:1::/48` local-use prefix) — were not decoded and checked against the private IPv4 ranges. - Redirects were validated only _after_ `fetch` had already followed them, so the request to a redirect target (e.g. an internal/metadata address) had already been issued before the check ran. - Several reserved/internal address ranges were not blocked: CGNAT (`100.64.0.0/10`, used by some cloud providers for internal traffic), benchmarking (`198.18.0.0/15`), IETF protocol assignments (`192.0.0.0/24`), the reserved `240.0.0.0/4` block (including the `255.255.255.255` broadcast address), and IPv6 site-local (`fec0::/10`) and multicast (`ff00::/8`). The validator now strips trailing dots before the hostname checks and fully expands IPv6 addresses to detect embedded private IPv4 targets. The download helpers now follow redirects manually (`redirect: 'manual'`), re-validating each hop before requesting it, so an unsafe redirect target is never fetched. When a redirect cannot be inspected because the runtime returns an opaque response, the helpers fail closed (reject the redirect) on the server; only in a real browser — where SSRF is not reachable (fetch is constrained by CORS and cannot reach a server's internal network or cloud-metadata endpoints) — is the redirect followed natively so legitimate redirected downloads keep working. - f18b08f: fix: redact server error details from UI message streams by default `toUIMessageStream`, `createUIMessageStream`, and `toUIMessageChunk` defaulted their `onError` callback to `getErrorMessage`, which serializes the raw error (`error.toString()` / `JSON.stringify(error)`) into the client-facing `{ type: 'error', errorText }` chunk — and also into `tool-output-error` parts. The documented default was `() => 'An error occurred.'`, so applications relying on the documented behavior were unknowingly streaming server exception details (internal hostnames, paths, provider request data, validation inputs) to end users. The default `onError` now returns the documented generic `'An error occurred.'`. Raw error details are only emitted when the developer explicitly supplies an `onError` handler. This also redacts `tool-output-error` and invalid-tool-input error text by default; pass an `onError` to surface richer messages. - b4507d5: fix(provider-utils): cancel response body on download rejection to prevent socket leak When a download was rejected early — because the `Content-Length` header exceeded the size limit, the response status was not ok, or a redirect resolved to a blocked URL — the fetch response body was left unconsumed and uncancelled. With WHATWG Fetch/undici this leaves the underlying TCP socket open instead of returning it to the connection pool, allowing an attacker-controlled origin to exhaust file descriptors and cause a denial of service. The body is now cancelled on all early-rejection paths in `readResponseWithSizeLimit`, `download`, and `downloadBlob`, and `fetchWithValidatedRedirects` cancels each redirect hop's body before following or rejecting the next hop. - Updated dependencies [8c17bf8] - Updated dependencies [aeda373] - Updated dependencies [558777f] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/gateway@4.0.0-canary.104 - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/alibaba@2.0.0-canary.59 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/amazon-bedrock@5.0.0-canary.84 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/anthropic@4.0.0-canary.66 - @ai-sdk/openai@4.0.0-canary.73 ## @ai-sdk/angular@3.0.0-canary.172 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/anthropic@4.0.0-canary.66 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/anthropic-aws@1.0.0-canary.7 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/anthropic@4.0.0-canary.66 ## @ai-sdk/assemblyai@3.0.0-canary.50 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/azure@4.0.0-canary.75 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/deepseek@3.0.0-canary.54 - @ai-sdk/openai@4.0.0-canary.73 ## @ai-sdk/baseten@2.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/black-forest-labs@2.0.0-canary.50 ### Patch Changes - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/bytedance@2.0.0-canary.51 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/cerebras@3.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/cohere@4.0.0-canary.52 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/deepgram@3.0.0-canary.49 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/deepinfra@3.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/deepseek@3.0.0-canary.54 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/devtools@1.0.0-canary.30 ### Patch Changes - 32ce45c: fix(devtools): secure devtools viewer api access - 275674e: fix(devtools): validate viewer database paths ## @ai-sdk/elevenlabs@3.0.0-canary.49 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/fal@3.0.0-canary.49 ### Patch Changes - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/fireworks@3.0.0-canary.56 ### Patch Changes - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/gateway@4.0.0-canary.104 ### Patch Changes - 8c17bf8: fix(gateway): surface provider warnings in embedding and reranking responses - 558777f: fix(gateway): accept deprecated warnings in image, speech, transcription, and video responses - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/gladia@3.0.0-canary.49 ### Patch Changes - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/google@4.0.0-canary.81 ### Patch Changes - 5878b40: fix(google): prevent prototype pollution when streaming tool args - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/google-vertex@5.0.0-canary.107 ### Patch Changes - Updated dependencies [5878b40] - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/google@4.0.0-canary.81 - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/anthropic@4.0.0-canary.66 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/groq@4.0.0-canary.53 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/harness@1.0.0-canary.7 ### Patch Changes - 3d87086: fix(harness): guard against invalid resuming a session vs continuing a turn - 1ea15a3: fix(harness): fix various bugs with harness skills not being correctly processed by the harness adapters - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/harness-claude-code@1.0.0-canary.3 ### Patch Changes - 1ea15a3: fix(harness): fix various bugs with harness skills not being correctly processed by the harness adapters - e551763: fix(harness): avoid using peer dependencies for underlying harness and sandbox SDKs - Updated dependencies [3d87086] - Updated dependencies [aeda373] - Updated dependencies [1ea15a3] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/harness@1.0.0-canary.7 - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/harness-codex@1.0.0-canary.3 ### Patch Changes - 1ea15a3: fix(harness): fix various bugs with harness skills not being correctly processed by the harness adapters - e551763: fix(harness): avoid using peer dependencies for underlying harness and sandbox SDKs - Updated dependencies [3d87086] - Updated dependencies [aeda373] - Updated dependencies [1ea15a3] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/harness@1.0.0-canary.7 - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/harness-pi@1.0.0-canary.3 ### Patch Changes - 1ea15a3: fix(harness): fix various bugs with harness skills not being correctly processed by the harness adapters - e551763: fix(harness): avoid using peer dependencies for underlying harness and sandbox SDKs - Updated dependencies [3d87086] - Updated dependencies [aeda373] - Updated dependencies [1ea15a3] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/harness@1.0.0-canary.7 - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/huggingface@2.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/hume@3.0.0-canary.49 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/klingai@4.0.0-canary.50 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/langchain@3.0.0-canary.172 ### Patch Changes - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - ai@7.0.0-canary.172 ## @ai-sdk/llamaindex@3.0.0-canary.172 ### Patch Changes - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - ai@7.0.0-canary.172 ## @ai-sdk/lmnt@3.0.0-canary.49 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/luma@3.0.0-canary.49 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/mcp@2.0.0-canary.63 ### Patch Changes - 2a150f8: fix(mcp): lock first sse endpoint received via event - b44b051: fix(mcp): prevent prototype-named tools from bypassing the `schemas` allowlist When using `client.tools({ schemas })` to expose only an explicitly allowed subset of an MCP server's tools, the allowlist check used the `in` operator, which also matches inherited `Object.prototype` properties. A server-advertised tool named `constructor`, `toString`, `__proto__`, etc. would pass the check even though the developer never defined it in `schemas`, and was then exposed to the model and executable. The check now uses `Object.hasOwn`, so only explicitly defined tools are returned. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/mistral@4.0.0-canary.54 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/moonshotai@3.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/open-responses@2.0.0-canary.53 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/openai@4.0.0-canary.73 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/openai-compatible@3.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/otel@1.0.0-canary.118 ### Patch Changes - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - ai@7.0.0-canary.172 ## @ai-sdk/perplexity@4.0.0-canary.52 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/policy-opa@1.0.0-canary.9 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/prodia@2.0.0-canary.52 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/provider-utils@5.0.0-canary.48 ### Patch Changes - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - 375fdd7: fix: harden download URL SSRF guard against hostname and redirect bypasses `validateDownloadUrl` and the file download helpers (`downloadBlob`, `download`) could be bypassed in several ways when handling untrusted URLs: - A fully-qualified hostname with a trailing dot (e.g. `localhost.`, `myhost.local.`) skipped the localhost/`.local` blocklist. - IPv6 addresses that embed an IPv4 address in their last 32 bits — IPv4-compatible (`::127.0.0.1`), IPv4-translated (`::ffff:0:127.0.0.1`), and NAT64 (`64:ff9b::127.0.0.1`, including the `64:ff9b:1::/48` local-use prefix) — were not decoded and checked against the private IPv4 ranges. - Redirects were validated only _after_ `fetch` had already followed them, so the request to a redirect target (e.g. an internal/metadata address) had already been issued before the check ran. - Several reserved/internal address ranges were not blocked: CGNAT (`100.64.0.0/10`, used by some cloud providers for internal traffic), benchmarking (`198.18.0.0/15`), IETF protocol assignments (`192.0.0.0/24`), the reserved `240.0.0.0/4` block (including the `255.255.255.255` broadcast address), and IPv6 site-local (`fec0::/10`) and multicast (`ff00::/8`). The validator now strips trailing dots before the hostname checks and fully expands IPv6 addresses to detect embedded private IPv4 targets. The download helpers now follow redirects manually (`redirect: 'manual'`), re-validating each hop before requesting it, so an unsafe redirect target is never fetched. When a redirect cannot be inspected because the runtime returns an opaque response, the helpers fail closed (reject the redirect) on the server; only in a real browser — where SSRF is not reachable (fetch is constrained by CORS and cannot reach a server's internal network or cloud-metadata endpoints) — is the redirect followed natively so legitimate redirected downloads keep working. - b4507d5: fix(provider-utils): cancel response body on download rejection to prevent socket leak When a download was rejected early — because the `Content-Length` header exceeded the size limit, the response status was not ok, or a redirect resolved to a blocked URL — the fetch response body was left unconsumed and uncancelled. With WHATWG Fetch/undici this leaves the underlying TCP socket open instead of returning it to the connection pool, allowing an attacker-controlled origin to exhaust file descriptors and cause a denial of service. The body is now cancelled on all early-rejection paths in `readResponseWithSizeLimit`, `download`, and `downloadBlob`, and `fetchWithValidatedRedirects` cancels each redirect hop's body before following or rejecting the next hop. ## @ai-sdk/quiverai@2.0.0-canary.5 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/react@4.0.0-canary.175 ### Patch Changes - 555c5de: fix(react): deny MCP App tool calls by default when allowedTools is omitted `experimental_MCPAppRenderer`'s bridge only enforced the `allowedTools` allowlist when it was non-null, so omitting `allowedTools` skipped the check and forwarded every `tools/call` from the (untrusted) MCP App iframe to the host's `callTool`. A malicious or compromised MCP server could therefore invoke any tool the host wired up. Tool invocation from MCP App content is now deny-by-default: if `allowedTools` is not explicitly provided, all `tools/call` requests are rejected. To expose tools to an app, list them in `handlers.allowedTools`. - Updated dependencies [2a150f8] - Updated dependencies [b44b051] - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/mcp@2.0.0-canary.63 - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/replicate@3.0.0-canary.50 ### Patch Changes - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/revai@3.0.0-canary.50 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/rsc@3.0.0-canary.173 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/sandbox-just-bash@1.0.0-canary.7 ### Patch Changes - e551763: fix(harness): avoid using peer dependencies for underlying harness and sandbox SDKs - Updated dependencies [3d87086] - Updated dependencies [aeda373] - Updated dependencies [1ea15a3] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/harness@1.0.0-canary.7 - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/sandbox-vercel@1.0.0-canary.7 ### Patch Changes - e551763: fix(harness): avoid using peer dependencies for underlying harness and sandbox SDKs - Updated dependencies [3d87086] - Updated dependencies [aeda373] - Updated dependencies [1ea15a3] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/harness@1.0.0-canary.7 - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/svelte@5.0.0-canary.172 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/togetherai@3.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/tui@1.0.0-canary.7 ### Patch Changes - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - ai@7.0.0-canary.172 ## @ai-sdk/valibot@3.0.0-canary.48 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/vercel@3.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/voyage@2.0.0-canary.23 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/vue@4.0.0-canary.172 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/workflow@1.0.0-canary.89 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/xai@4.0.0-canary.74 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Version Packages (canary) (vercel#16084) This PR was opened by the [Changesets release](https://github.com/changesets/action) GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to main, this PR will be updated.⚠️ ⚠️ ⚠️ ⚠️ ⚠️ ⚠️ `main` is currently in **pre mode** so this branch has prereleases rather than normal releases. If you want to exit prereleases, run `changeset pre exit` on `main`.⚠️ ⚠️ ⚠️ ⚠️ ⚠️ ⚠️ # Releases ## ai@7.0.0-canary.173 ### Patch Changes - Updated dependencies [efec111] - @ai-sdk/gateway@4.0.0-canary.105 ## @ai-sdk/angular@3.0.0-canary.173 ### Patch Changes - ai@7.0.0-canary.173 ## @ai-sdk/fireworks@3.0.0-canary.57 ### Patch Changes - 38010a1: Enable `includeUsage` for Fireworks so streaming responses report token usage ## @ai-sdk/gateway@4.0.0-canary.105 ### Patch Changes - efec111: chore(provider/gateway): update gateway model settings files ## @ai-sdk/harness@1.0.0-canary.9 ### Patch Changes - ai@7.0.0-canary.173 ## @ai-sdk/harness-claude-code@1.0.0-canary.5 ### Patch Changes - @ai-sdk/harness@1.0.0-canary.9 ## @ai-sdk/harness-codex@1.0.0-canary.5 ### Patch Changes - @ai-sdk/harness@1.0.0-canary.9 ## @ai-sdk/harness-pi@1.0.0-canary.5 ### Patch Changes - @ai-sdk/harness@1.0.0-canary.9 ## @ai-sdk/langchain@3.0.0-canary.173 ### Patch Changes - c1afaed: fix(langchain): prevent polluting global object.prototype - ai@7.0.0-canary.173 ## @ai-sdk/llamaindex@3.0.0-canary.173 ### Patch Changes - ai@7.0.0-canary.173 ## @ai-sdk/mcp@2.0.0-canary.64 ### Patch Changes - 024a6b4: fix(mcp): validate oauth metadata issuer during discovery ## @ai-sdk/otel@1.0.0-canary.119 ### Patch Changes - ai@7.0.0-canary.173 ## @ai-sdk/policy-opa@1.0.0-canary.10 ### Patch Changes - ai@7.0.0-canary.173 ## @ai-sdk/react@4.0.0-canary.176 ### Patch Changes - Updated dependencies [024a6b4] - @ai-sdk/mcp@2.0.0-canary.64 - ai@7.0.0-canary.173 ## @ai-sdk/rsc@3.0.0-canary.174 ### Patch Changes - ai@7.0.0-canary.173 ## @ai-sdk/sandbox-just-bash@1.0.0-canary.9 ### Patch Changes - @ai-sdk/harness@1.0.0-canary.9 ## @ai-sdk/sandbox-vercel@1.0.0-canary.9 ### Patch Changes - @ai-sdk/harness@1.0.0-canary.9 ## @ai-sdk/svelte@5.0.0-canary.173 ### Patch Changes - ai@7.0.0-canary.173 ## @ai-sdk/tui@1.0.0-canary.8 ### Patch Changes - ai@7.0.0-canary.173 ## @ai-sdk/vue@4.0.0-canary.173 ### Patch Changes - ai@7.0.0-canary.173 ## @ai-sdk/workflow@1.0.0-canary.90 ### Patch Changes - ai@7.0.0-canary.173 Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Version Packages (canary) (vercel#16024) This PR was opened by the [Changesets release](https://github.com/changesets/action) GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to main, this PR will be updated.⚠️ ⚠️ ⚠️ ⚠️ ⚠️ ⚠️ `main` is currently in **pre mode** so this branch has prereleases rather than normal releases. If you want to exit prereleases, run `changeset pre exit` on `main`.⚠️ ⚠️ ⚠️ ⚠️ ⚠️ ⚠️ # Releases ## ai@7.0.0-canary.172 ### Patch Changes - 25a64f8: Remove deprecated experimental generateImage exports. - 375fdd7: fix: harden download URL SSRF guard against hostname and redirect bypasses `validateDownloadUrl` and the file download helpers (`downloadBlob`, `download`) could be bypassed in several ways when handling untrusted URLs: - A fully-qualified hostname with a trailing dot (e.g. `localhost.`, `myhost.local.`) skipped the localhost/`.local` blocklist. - IPv6 addresses that embed an IPv4 address in their last 32 bits — IPv4-compatible (`::127.0.0.1`), IPv4-translated (`::ffff:0:127.0.0.1`), and NAT64 (`64:ff9b::127.0.0.1`, including the `64:ff9b:1::/48` local-use prefix) — were not decoded and checked against the private IPv4 ranges. - Redirects were validated only _after_ `fetch` had already followed them, so the request to a redirect target (e.g. an internal/metadata address) had already been issued before the check ran. - Several reserved/internal address ranges were not blocked: CGNAT (`100.64.0.0/10`, used by some cloud providers for internal traffic), benchmarking (`198.18.0.0/15`), IETF protocol assignments (`192.0.0.0/24`), the reserved `240.0.0.0/4` block (including the `255.255.255.255` broadcast address), and IPv6 site-local (`fec0::/10`) and multicast (`ff00::/8`). The validator now strips trailing dots before the hostname checks and fully expands IPv6 addresses to detect embedded private IPv4 targets. The download helpers now follow redirects manually (`redirect: 'manual'`), re-validating each hop before requesting it, so an unsafe redirect target is never fetched. When a redirect cannot be inspected because the runtime returns an opaque response, the helpers fail closed (reject the redirect) on the server; only in a real browser — where SSRF is not reachable (fetch is constrained by CORS and cannot reach a server's internal network or cloud-metadata endpoints) — is the redirect followed natively so legitimate redirected downloads keep working. - f18b08f: fix: redact server error details from UI message streams by default `toUIMessageStream`, `createUIMessageStream`, and `toUIMessageChunk` defaulted their `onError` callback to `getErrorMessage`, which serializes the raw error (`error.toString()` / `JSON.stringify(error)`) into the client-facing `{ type: 'error', errorText }` chunk — and also into `tool-output-error` parts. The documented default was `() => 'An error occurred.'`, so applications relying on the documented behavior were unknowingly streaming server exception details (internal hostnames, paths, provider request data, validation inputs) to end users. The default `onError` now returns the documented generic `'An error occurred.'`. Raw error details are only emitted when the developer explicitly supplies an `onError` handler. This also redacts `tool-output-error` and invalid-tool-input error text by default; pass an `onError` to surface richer messages. - b4507d5: fix(provider-utils): cancel response body on download rejection to prevent socket leak When a download was rejected early — because the `Content-Length` header exceeded the size limit, the response status was not ok, or a redirect resolved to a blocked URL — the fetch response body was left unconsumed and uncancelled. With WHATWG Fetch/undici this leaves the underlying TCP socket open instead of returning it to the connection pool, allowing an attacker-controlled origin to exhaust file descriptors and cause a denial of service. The body is now cancelled on all early-rejection paths in `readResponseWithSizeLimit`, `download`, and `downloadBlob`, and `fetchWithValidatedRedirects` cancels each redirect hop's body before following or rejecting the next hop. - Updated dependencies [8c17bf8] - Updated dependencies [aeda373] - Updated dependencies [558777f] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/gateway@4.0.0-canary.104 - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/alibaba@2.0.0-canary.59 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/amazon-bedrock@5.0.0-canary.84 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/anthropic@4.0.0-canary.66 - @ai-sdk/openai@4.0.0-canary.73 ## @ai-sdk/angular@3.0.0-canary.172 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/anthropic@4.0.0-canary.66 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/anthropic-aws@1.0.0-canary.7 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/anthropic@4.0.0-canary.66 ## @ai-sdk/assemblyai@3.0.0-canary.50 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/azure@4.0.0-canary.75 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/deepseek@3.0.0-canary.54 - @ai-sdk/openai@4.0.0-canary.73 ## @ai-sdk/baseten@2.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/black-forest-labs@2.0.0-canary.50 ### Patch Changes - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/bytedance@2.0.0-canary.51 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/cerebras@3.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/cohere@4.0.0-canary.52 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/deepgram@3.0.0-canary.49 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/deepinfra@3.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/deepseek@3.0.0-canary.54 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/devtools@1.0.0-canary.30 ### Patch Changes - 32ce45c: fix(devtools): secure devtools viewer api access - 275674e: fix(devtools): validate viewer database paths ## @ai-sdk/elevenlabs@3.0.0-canary.49 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/fal@3.0.0-canary.49 ### Patch Changes - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/fireworks@3.0.0-canary.56 ### Patch Changes - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/gateway@4.0.0-canary.104 ### Patch Changes - 8c17bf8: fix(gateway): surface provider warnings in embedding and reranking responses - 558777f: fix(gateway): accept deprecated warnings in image, speech, transcription, and video responses - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/gladia@3.0.0-canary.49 ### Patch Changes - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/google@4.0.0-canary.81 ### Patch Changes - 5878b40: fix(google): prevent prototype pollution when streaming tool args - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/google-vertex@5.0.0-canary.107 ### Patch Changes - Updated dependencies [5878b40] - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/google@4.0.0-canary.81 - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/anthropic@4.0.0-canary.66 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/groq@4.0.0-canary.53 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/harness@1.0.0-canary.7 ### Patch Changes - 3d87086: fix(harness): guard against invalid resuming a session vs continuing a turn - 1ea15a3: fix(harness): fix various bugs with harness skills not being correctly processed by the harness adapters - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/harness-claude-code@1.0.0-canary.3 ### Patch Changes - 1ea15a3: fix(harness): fix various bugs with harness skills not being correctly processed by the harness adapters - e551763: fix(harness): avoid using peer dependencies for underlying harness and sandbox SDKs - Updated dependencies [3d87086] - Updated dependencies [aeda373] - Updated dependencies [1ea15a3] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/harness@1.0.0-canary.7 - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/harness-codex@1.0.0-canary.3 ### Patch Changes - 1ea15a3: fix(harness): fix various bugs with harness skills not being correctly processed by the harness adapters - e551763: fix(harness): avoid using peer dependencies for underlying harness and sandbox SDKs - Updated dependencies [3d87086] - Updated dependencies [aeda373] - Updated dependencies [1ea15a3] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/harness@1.0.0-canary.7 - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/harness-pi@1.0.0-canary.3 ### Patch Changes - 1ea15a3: fix(harness): fix various bugs with harness skills not being correctly processed by the harness adapters - e551763: fix(harness): avoid using peer dependencies for underlying harness and sandbox SDKs - Updated dependencies [3d87086] - Updated dependencies [aeda373] - Updated dependencies [1ea15a3] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/harness@1.0.0-canary.7 - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/huggingface@2.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/hume@3.0.0-canary.49 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/klingai@4.0.0-canary.50 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/langchain@3.0.0-canary.172 ### Patch Changes - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - ai@7.0.0-canary.172 ## @ai-sdk/llamaindex@3.0.0-canary.172 ### Patch Changes - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - ai@7.0.0-canary.172 ## @ai-sdk/lmnt@3.0.0-canary.49 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/luma@3.0.0-canary.49 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/mcp@2.0.0-canary.63 ### Patch Changes - 2a150f8: fix(mcp): lock first sse endpoint received via event - b44b051: fix(mcp): prevent prototype-named tools from bypassing the `schemas` allowlist When using `client.tools({ schemas })` to expose only an explicitly allowed subset of an MCP server's tools, the allowlist check used the `in` operator, which also matches inherited `Object.prototype` properties. A server-advertised tool named `constructor`, `toString`, `__proto__`, etc. would pass the check even though the developer never defined it in `schemas`, and was then exposed to the model and executable. The check now uses `Object.hasOwn`, so only explicitly defined tools are returned. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/mistral@4.0.0-canary.54 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/moonshotai@3.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/open-responses@2.0.0-canary.53 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/openai@4.0.0-canary.73 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/openai-compatible@3.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/otel@1.0.0-canary.118 ### Patch Changes - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - ai@7.0.0-canary.172 ## @ai-sdk/perplexity@4.0.0-canary.52 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/policy-opa@1.0.0-canary.9 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/prodia@2.0.0-canary.52 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/provider-utils@5.0.0-canary.48 ### Patch Changes - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - 375fdd7: fix: harden download URL SSRF guard against hostname and redirect bypasses `validateDownloadUrl` and the file download helpers (`downloadBlob`, `download`) could be bypassed in several ways when handling untrusted URLs: - A fully-qualified hostname with a trailing dot (e.g. `localhost.`, `myhost.local.`) skipped the localhost/`.local` blocklist. - IPv6 addresses that embed an IPv4 address in their last 32 bits — IPv4-compatible (`::127.0.0.1`), IPv4-translated (`::ffff:0:127.0.0.1`), and NAT64 (`64:ff9b::127.0.0.1`, including the `64:ff9b:1::/48` local-use prefix) — were not decoded and checked against the private IPv4 ranges. - Redirects were validated only _after_ `fetch` had already followed them, so the request to a redirect target (e.g. an internal/metadata address) had already been issued before the check ran. - Several reserved/internal address ranges were not blocked: CGNAT (`100.64.0.0/10`, used by some cloud providers for internal traffic), benchmarking (`198.18.0.0/15`), IETF protocol assignments (`192.0.0.0/24`), the reserved `240.0.0.0/4` block (including the `255.255.255.255` broadcast address), and IPv6 site-local (`fec0::/10`) and multicast (`ff00::/8`). The validator now strips trailing dots before the hostname checks and fully expands IPv6 addresses to detect embedded private IPv4 targets. The download helpers now follow redirects manually (`redirect: 'manual'`), re-validating each hop before requesting it, so an unsafe redirect target is never fetched. When a redirect cannot be inspected because the runtime returns an opaque response, the helpers fail closed (reject the redirect) on the server; only in a real browser — where SSRF is not reachable (fetch is constrained by CORS and cannot reach a server's internal network or cloud-metadata endpoints) — is the redirect followed natively so legitimate redirected downloads keep working. - b4507d5: fix(provider-utils): cancel response body on download rejection to prevent socket leak When a download was rejected early — because the `Content-Length` header exceeded the size limit, the response status was not ok, or a redirect resolved to a blocked URL — the fetch response body was left unconsumed and uncancelled. With WHATWG Fetch/undici this leaves the underlying TCP socket open instead of returning it to the connection pool, allowing an attacker-controlled origin to exhaust file descriptors and cause a denial of service. The body is now cancelled on all early-rejection paths in `readResponseWithSizeLimit`, `download`, and `downloadBlob`, and `fetchWithValidatedRedirects` cancels each redirect hop's body before following or rejecting the next hop. ## @ai-sdk/quiverai@2.0.0-canary.5 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/react@4.0.0-canary.175 ### Patch Changes - 555c5de: fix(react): deny MCP App tool calls by default when allowedTools is omitted `experimental_MCPAppRenderer`'s bridge only enforced the `allowedTools` allowlist when it was non-null, so omitting `allowedTools` skipped the check and forwarded every `tools/call` from the (untrusted) MCP App iframe to the host's `callTool`. A malicious or compromised MCP server could therefore invoke any tool the host wired up. Tool invocation from MCP App content is now deny-by-default: if `allowedTools` is not explicitly provided, all `tools/call` requests are rejected. To expose tools to an app, list them in `handlers.allowedTools`. - Updated dependencies [2a150f8] - Updated dependencies [b44b051] - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/mcp@2.0.0-canary.63 - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/replicate@3.0.0-canary.50 ### Patch Changes - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/revai@3.0.0-canary.50 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/rsc@3.0.0-canary.173 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/sandbox-just-bash@1.0.0-canary.7 ### Patch Changes - e551763: fix(harness): avoid using peer dependencies for underlying harness and sandbox SDKs - Updated dependencies [3d87086] - Updated dependencies [aeda373] - Updated dependencies [1ea15a3] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/harness@1.0.0-canary.7 - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/sandbox-vercel@1.0.0-canary.7 ### Patch Changes - e551763: fix(harness): avoid using peer dependencies for underlying harness and sandbox SDKs - Updated dependencies [3d87086] - Updated dependencies [aeda373] - Updated dependencies [1ea15a3] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/harness@1.0.0-canary.7 - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/svelte@5.0.0-canary.172 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/togetherai@3.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/tui@1.0.0-canary.7 ### Patch Changes - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - ai@7.0.0-canary.172 ## @ai-sdk/valibot@3.0.0-canary.48 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/vercel@3.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/voyage@2.0.0-canary.23 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/vue@4.0.0-canary.172 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/workflow@1.0.0-canary.89 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/xai@4.0.0-canary.74 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Version Packages (vercel#16026) This PR was opened by the [Changesets release](https://github.com/changesets/action) GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to release-v6.0, this PR will be updated. # Releases ## ai@6.0.203 ### Patch Changes - f42aa79: fix: harden download URL SSRF guard against hostname and redirect bypasses `validateDownloadUrl` and the file download helpers (`downloadBlob`, `download`) could be bypassed in several ways when handling untrusted URLs: - A fully-qualified hostname with a trailing dot (e.g. `localhost.`, `myhost.local.`) skipped the localhost/`.local` blocklist. - IPv6 addresses that embed an IPv4 address in their last 32 bits — IPv4-compatible (`::127.0.0.1`), IPv4-translated (`::ffff:0:127.0.0.1`), and NAT64 (`64:ff9b::127.0.0.1`, including the `64:ff9b:1::/48` local-use prefix) — were not decoded and checked against the private IPv4 ranges. - Redirects were validated only _after_ `fetch` had already followed them, so the request to a redirect target (e.g. an internal/metadata address) had already been issued before the check ran. - Several reserved/internal address ranges were not blocked: CGNAT (`100.64.0.0/10`, used by some cloud providers for internal traffic), benchmarking (`198.18.0.0/15`), IETF protocol assignments (`192.0.0.0/24`), the reserved `240.0.0.0/4` block (including the `255.255.255.255` broadcast address), and IPv6 site-local (`fec0::/10`) and multicast (`ff00::/8`). The validator now strips trailing dots before the hostname checks and fully expands IPv6 addresses to detect embedded private IPv4 targets. The download helpers now follow redirects manually (`redirect: 'manual'`), re-validating each hop before requesting it, so an unsafe redirect target is never fetched. When a redirect cannot be inspected because the runtime returns an opaque response, the helpers fail closed (reject the redirect) on the server; only in a real browser — where SSRF is not reachable (fetch is constrained by CORS and cannot reach a server's internal network or cloud-metadata endpoints) — is the redirect followed natively so legitimate redirected downloads keep working. - 5291f7e: Harden stream text processing and middleware against prototype pollution from stream part IDs. - b4b575a: fix: redact server error details from UI message streams by default `streamText(...).toUIMessageStream()` and `createUIMessageStream` defaulted their `onError` callback to `getErrorMessage`, which serializes the raw error (`error.toString()` / `JSON.stringify(error)`) into the client-facing `{ type: 'error', errorText }` chunk — and also into `tool-output-error` parts. The documented default was `() => 'An error occurred.'`, so applications relying on the documented behavior were unknowingly streaming server exception details (internal hostnames, paths, provider request data, validation inputs) to end users. The default `onError` now returns the documented generic `'An error occurred.'`. Raw error details are only emitted when the developer explicitly supplies an `onError` handler. This also redacts `tool-output-error` and invalid-tool-input error text by default; pass an `onError` to surface richer messages. - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/gateway@3.0.129 ## @ai-sdk/alibaba@1.0.28 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/openai-compatible@2.0.50 ## @ai-sdk/amazon-bedrock@4.0.117 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/anthropic@3.0.84 - @ai-sdk/openai@3.0.71 ## @ai-sdk/angular@2.0.204 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - Updated dependencies [5291f7e] - Updated dependencies [b4b575a] - @ai-sdk/provider-utils@4.0.29 - ai@6.0.203 ## @ai-sdk/anthropic@3.0.84 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/anthropic-aws@1.0.6 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/anthropic@3.0.84 ## @ai-sdk/assemblyai@2.0.35 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/azure@3.0.74 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/deepseek@2.0.38 - @ai-sdk/openai@3.0.71 ## @ai-sdk/baseten@1.0.53 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/openai-compatible@2.0.50 ## @ai-sdk/black-forest-labs@1.0.36 ### Patch Changes - bfa5864: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/bytedance@1.0.17 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/cerebras@2.0.56 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/openai-compatible@2.0.50 ## @ai-sdk/cohere@3.0.38 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/deepgram@2.0.35 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/deepinfra@2.0.54 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/openai-compatible@2.0.50 ## @ai-sdk/deepseek@2.0.38 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/devtools@0.0.19 ### Patch Changes - f27afa8: fix(devtools): secure devtools viewer api access ## @ai-sdk/elevenlabs@2.0.35 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/fal@2.0.36 ### Patch Changes - bfa5864: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/fireworks@2.0.55 ### Patch Changes - bfa5864: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/openai-compatible@2.0.50 ## @ai-sdk/gateway@3.0.129 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/gladia@2.0.35 ### Patch Changes - bfa5864: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/google@3.0.82 ### Patch Changes - 3258f22: fix(google): prevent prototype pollution when streaming tool args - bfa5864: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/google-vertex@4.0.145 ### Patch Changes - Updated dependencies [3258f22] - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/google@3.0.82 - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/anthropic@3.0.84 - @ai-sdk/openai-compatible@2.0.50 ## @ai-sdk/groq@3.0.41 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/huggingface@1.0.52 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/openai-compatible@2.0.50 ## @ai-sdk/hume@2.0.35 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/klingai@3.0.20 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/langchain@2.0.210 ### Patch Changes - Updated dependencies [f42aa79] - Updated dependencies [5291f7e] - Updated dependencies [b4b575a] - ai@6.0.203 ## @ai-sdk/llamaindex@2.0.203 ### Patch Changes - Updated dependencies [f42aa79] - Updated dependencies [5291f7e] - Updated dependencies [b4b575a] - ai@6.0.203 ## @ai-sdk/lmnt@2.0.35 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/luma@2.0.35 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/mcp@1.0.49 ### Patch Changes - 3e8d9ba: fix(mcp): lock first sse endpoint received via event - 4fa7354: fix(mcp): prevent prototype-named tools from bypassing the `schemas` allowlist When using `client.tools({ schemas })` to expose only an explicitly allowed subset of an MCP server's tools, the allowlist check used the `in` operator, which also matches inherited `Object.prototype` properties. A server-advertised tool named `constructor`, `toString`, `__proto__`, etc. would pass the check even though the developer never defined it in `schemas`, and was then exposed to the model and executable. The check now uses `Object.hasOwn`, so only explicitly defined tools are returned. - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/mistral@3.0.39 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/moonshotai@2.0.25 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/openai-compatible@2.0.50 ## @ai-sdk/open-responses@1.0.18 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/openai@3.0.71 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/openai-compatible@2.0.50 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/perplexity@3.0.35 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/prodia@1.0.34 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/provider-utils@4.0.29 ### Patch Changes - bfa5864: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - f42aa79: fix: harden download URL SSRF guard against hostname and redirect bypasses `validateDownloadUrl` and the file download helpers (`downloadBlob`, `download`) could be bypassed in several ways when handling untrusted URLs: - A fully-qualified hostname with a trailing dot (e.g. `localhost.`, `myhost.local.`) skipped the localhost/`.local` blocklist. - IPv6 addresses that embed an IPv4 address in their last 32 bits — IPv4-compatible (`::127.0.0.1`), IPv4-translated (`::ffff:0:127.0.0.1`), and NAT64 (`64:ff9b::127.0.0.1`, including the `64:ff9b:1::/48` local-use prefix) — were not decoded and checked against the private IPv4 ranges. - Redirects were validated only _after_ `fetch` had already followed them, so the request to a redirect target (e.g. an internal/metadata address) had already been issued before the check ran. - Several reserved/internal address ranges were not blocked: CGNAT (`100.64.0.0/10`, used by some cloud providers for internal traffic), benchmarking (`198.18.0.0/15`), IETF protocol assignments (`192.0.0.0/24`), the reserved `240.0.0.0/4` block (including the `255.255.255.255` broadcast address), and IPv6 site-local (`fec0::/10`) and multicast (`ff00::/8`). The validator now strips trailing dots before the hostname checks and fully expands IPv6 addresses to detect embedded private IPv4 targets. The download helpers now follow redirects manually (`redirect: 'manual'`), re-validating each hop before requesting it, so an unsafe redirect target is never fetched. When a redirect cannot be inspected because the runtime returns an opaque response, the helpers fail closed (reject the redirect) on the server; only in a real browser — where SSRF is not reachable (fetch is constrained by CORS and cannot reach a server's internal network or cloud-metadata endpoints) — is the redirect followed natively so legitimate redirected downloads keep working. ## @ai-sdk/quiverai@1.0.2 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/react@3.0.205 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - Updated dependencies [5291f7e] - Updated dependencies [b4b575a] - @ai-sdk/provider-utils@4.0.29 - ai@6.0.203 ## @ai-sdk/replicate@2.0.35 ### Patch Changes - bfa5864: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/revai@2.0.35 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/rsc@2.0.203 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - Updated dependencies [5291f7e] - Updated dependencies [b4b575a] - @ai-sdk/provider-utils@4.0.29 - ai@6.0.203 ## @ai-sdk/svelte@4.0.203 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - Updated dependencies [5291f7e] - Updated dependencies [b4b575a] - @ai-sdk/provider-utils@4.0.29 - ai@6.0.203 ## @ai-sdk/togetherai@2.0.55 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/openai-compatible@2.0.50 ## @ai-sdk/valibot@2.0.30 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/vercel@2.0.52 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/openai-compatible@2.0.50 ## @ai-sdk/voyage@1.0.6 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 ## @ai-sdk/vue@3.0.203 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - Updated dependencies [5291f7e] - Updated dependencies [b4b575a] - @ai-sdk/provider-utils@4.0.29 - ai@6.0.203 ## @ai-sdk/xai@3.0.95 ### Patch Changes - Updated dependencies [bfa5864] - Updated dependencies [f42aa79] - @ai-sdk/provider-utils@4.0.29 - @ai-sdk/openai-compatible@2.0.50 Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Version Packages (vercel#16027) This PR was opened by the [Changesets release](https://github.com/changesets/action) GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to release-v5.0, this PR will be updated. # Releases ## ai@5.0.200 ### Patch Changes - eea9166: fix: harden download URL SSRF guard against hostname and redirect bypasses `validateDownloadUrl` and the file download helpers (`downloadBlob`, `download`) could be bypassed in several ways when handling untrusted URLs: - A fully-qualified hostname with a trailing dot (e.g. `localhost.`, `myhost.local.`) skipped the localhost/`.local` blocklist. - IPv6 addresses that embed an IPv4 address in their last 32 bits — IPv4-compatible (`::127.0.0.1`), IPv4-translated (`::ffff:0:127.0.0.1`), and NAT64 (`64:ff9b::127.0.0.1`, including the `64:ff9b:1::/48` local-use prefix) — were not decoded and checked against the private IPv4 ranges. - Redirects were validated only _after_ `fetch` had already followed them, so the request to a redirect target (e.g. an internal/metadata address) had already been issued before the check ran. - Several reserved/internal address ranges were not blocked: CGNAT (`100.64.0.0/10`, used by some cloud providers for internal traffic), benchmarking (`198.18.0.0/15`), IETF protocol assignments (`192.0.0.0/24`), the reserved `240.0.0.0/4` block (including the `255.255.255.255` broadcast address), and IPv6 site-local (`fec0::/10`) and multicast (`ff00::/8`). The validator now strips trailing dots before the hostname checks and fully expands IPv6 addresses to detect embedded private IPv4 targets. The download helpers now follow redirects manually (`redirect: 'manual'`), re-validating each hop before requesting it, so an unsafe redirect target is never fetched. When a redirect cannot be inspected because the runtime returns an opaque response, the helpers fail closed (reject the redirect) on the server; only in a real browser — where SSRF is not reachable (fetch is constrained by CORS and cannot reach a server's internal network or cloud-metadata endpoints) — is the redirect followed natively so legitimate redirected downloads keep working. - 15cf74f: Harden stream text processing and middleware against prototype pollution from stream part IDs. - 8ad6f80: fix: redact server error details from UI message streams by default `streamText(...).toUIMessageStream()` and `createUIMessageStream` defaulted their `onError` callback to `getErrorMessage`, which serializes the raw error (`error.toString()` / `JSON.stringify(error)`) into the client-facing `{ type: 'error', errorText }` chunk — and also into `tool-output-error` parts. The documented default was `() => 'An error occurred.'`, so applications relying on the documented behavior were unknowingly streaming server exception details (internal hostnames, paths, provider request data, validation inputs) to end users. The default `onError` now returns the documented generic `'An error occurred.'`. Raw error details are only emitted when the developer explicitly supplies an `onError` handler. This also redacts `tool-output-error` and invalid-tool-input error text by default; pass an `onError` to surface richer messages. - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/gateway@2.0.99 ## @ai-sdk/alibaba@0.0.12 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/openai-compatible@1.0.40 ## @ai-sdk/amazon-bedrock@3.0.102 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/anthropic@2.0.82 ## @ai-sdk/angular@1.0.200 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - Updated dependencies [15cf74f] - Updated dependencies [8ad6f80] - @ai-sdk/provider-utils@3.0.26 - ai@5.0.200 ## @ai-sdk/anthropic@2.0.82 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/assemblyai@1.0.30 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/azure@2.0.111 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/deepseek@1.0.43 - @ai-sdk/openai@2.0.107 ## @ai-sdk/baseten@0.0.23 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/openai-compatible@1.0.40 ## @ai-sdk/black-forest-labs@0.0.15 ### Patch Changes - 9f67efe: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/cerebras@1.0.45 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/openai-compatible@1.0.40 ## @ai-sdk/cohere@2.0.30 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/deepgram@1.0.29 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/deepinfra@1.0.43 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/openai-compatible@1.0.40 ## @ai-sdk/deepseek@1.0.43 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/elevenlabs@1.0.29 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/fal@1.0.33 ### Patch Changes - 9f67efe: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/fireworks@1.0.41 ### Patch Changes - 9f67efe: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/openai-compatible@1.0.40 ## @ai-sdk/gateway@2.0.99 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/gladia@1.0.28 ### Patch Changes - 9f67efe: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/google@2.0.75 ### Patch Changes - 9f67efe: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/google-vertex@3.0.142 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/google@2.0.75 - @ai-sdk/anthropic@2.0.82 - @ai-sdk/openai-compatible@1.0.40 ## @ai-sdk/groq@2.0.41 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/huggingface@0.0.23 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/openai-compatible@1.0.40 ## @ai-sdk/hume@1.0.28 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/langchain@1.0.200 ### Patch Changes - Updated dependencies [eea9166] - Updated dependencies [15cf74f] - Updated dependencies [8ad6f80] - ai@5.0.200 ## @ai-sdk/llamaindex@1.0.200 ### Patch Changes - Updated dependencies [eea9166] - Updated dependencies [15cf74f] - Updated dependencies [8ad6f80] - ai@5.0.200 ## @ai-sdk/lmnt@1.0.28 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/luma@1.0.28 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/mcp@0.0.19 ### Patch Changes - d872a7a: fix(mcp): lock first sse endpoint received via event - f4cd468: fix(mcp): prevent prototype-named tools from bypassing the `schemas` allowlist When using `client.tools({ schemas })` to expo 10BC8 se only an explicitly allowed subset of an MCP server's tools, the allowlist check used the `in` operator, which also matches inherited `Object.prototype` properties. A server-advertised tool named `constructor`, `toString`, `__proto__`, etc. would pass the check even though the developer never defined it in `schemas`, and was then exposed to the model and executable. The check now uses `Object.hasOwn`, so only explicitly defined tools are returned. - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/mistral@2.0.34 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/moonshotai@0.0.11 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/openai-compatible@1.0.40 ## @ai-sdk/openai@2.0.107 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/openai-compatible@1.0.40 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/perplexity@2.0.31 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/prodia@0.0.10 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/provider-utils@3.0.26 ### Patch Changes - 9f67efe: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - eea9166: fix: harden download URL SSRF guard against hostname and redirect bypasses `validateDownloadUrl` and the file download helpers (`downloadBlob`, `download`) could be bypassed in several ways when handling untrusted URLs: - A fully-qualified hostname with a trailing dot (e.g. `localhost.`, `myhost.local.`) skipped the localhost/`.local` blocklist. - IPv6 addresses that embed an IPv4 address in their last 32 bits — IPv4-compatible (`::127.0.0.1`), IPv4-translated (`::ffff:0:127.0.0.1`), and NAT64 (`64:ff9b::127.0.0.1`, including the `64:ff9b:1::/48` local-use prefix) — were not decoded and checked against the private IPv4 ranges. - Redirects were validated only _after_ `fetch` had already followed them, so the request to a redirect target (e.g. an internal/metadata address) had already been issued before the check ran. - Several reserved/internal address ranges were not blocked: CGNAT (`100.64.0.0/10`, used by some cloud providers for internal traffic), benchmarking (`198.18.0.0/15`), IETF protocol assignments (`192.0.0.0/24`), the reserved `240.0.0.0/4` block (including the `255.255.255.255` broadcast address), and IPv6 site-local (`fec0::/10`) and multicast (`ff00::/8`). The validator now strips trailing dots before the hostname checks and fully expands IPv6 addresses to detect embedded private IPv4 targets. The download helpers now follow redirects manually (`redirect: 'manual'`), re-validating each hop before requesting it, so an unsafe redirect target is never fetched. When a redirect cannot be inspected because the runtime returns an opaque response, the helpers fail closed (reject the redirect) on the server; only in a real browser — where SSRF is not reachable (fetch is constrained by CORS and cannot reach a server's internal network or cloud-metadata endpoints) — is the redirect followed natively so legitimate redirected downloads keep working. ## @ai-sdk/react@2.0.202 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - Updated dependencies [15cf74f] - Updated dependencies [8ad6f80] - @ai-sdk/provider-utils@3.0.26 - ai@5.0.200 ## @ai-sdk/replicate@1.0.28 ### Patch Changes - 9f67efe: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/revai@1.0.28 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/rsc@1.0.202 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - Updated dependencies [15cf74f] - Updated dependencies [8ad6f80] - @ai-sdk/provider-utils@3.0.26 - ai@5.0.200 ## @ai-sdk/svelte@3.0.200 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - Updated dependencies [15cf74f] - Updated dependencies [8ad6f80] - @ai-sdk/provider-utils@3.0.26 - ai@5.0.200 ## @ai-sdk/togetherai@1.0.43 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/openai-compatible@1.0.40 ## @ai-sdk/valibot@1.0.26 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 ## @ai-sdk/vercel@1.0.41 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/openai-compatible@1.0.40 ## @ai-sdk/vue@2.0.200 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - Updated dependencies [15cf74f] - Updated dependencies [8ad6f80] - @ai-sdk/provider-utils@3.0.26 - ai@5.0.200 ## @ai-sdk/xai@2.0.74 ### Patch Changes - Updated dependencies [9f67efe] - Updated dependencies [eea9166] - @ai-sdk/provider-utils@3.0.26 - @ai-sdk/openai-compatible@1.0.40 Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Version Packages (canary) (vercel#16024) This PR was opened by the [Changesets release](https://github.com/changesets/action) GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to main, this PR will be updated.⚠️ ⚠️ ⚠️ ⚠️ ⚠️ ⚠️ `main` is currently in **pre mode** so this branch has prereleases rather than normal releases. If you want to exit prereleases, run `changeset pre exit` on `main`.⚠️ ⚠️ ⚠️ ⚠️ ⚠️ ⚠️ # Releases ## ai@7.0.0-canary.172 ### Patch Changes - 25a64f8: Remove deprecated experimental generateImage exports. - 375fdd7: fix: harden download URL SSRF guard against hostname and redirect bypasses `validateDownloadUrl` and the file download helpers (`downloadBlob`, `download`) could be bypassed in several ways when handling untrusted URLs: - A fully-qualified hostname with a trailing dot (e.g. `localhost.`, `myhost.local.`) skipped the localhost/`.local` blocklist. - IPv6 addresses that embed an IPv4 address in their last 32 bits — IPv4-compatible (`::127.0.0.1`), IPv4-translated (`::ffff:0:127.0.0.1`), and NAT64 (`64:ff9b::127.0.0.1`, including the `64:ff9b:1::/48` local-use prefix) — were not decoded and checked against the private IPv4 ranges. - Redirects were validated only _after_ `fetch` had already followed them, so the request to a redirect target (e.g. an internal/metadata address) had already been issued before the check ran. - Several reserved/internal address ranges were not blocked: CGNAT (`100.64.0.0/10`, used by some cloud providers for internal traffic), benchmarking (`198.18.0.0/15`), IETF protocol assignments (`192.0.0.0/24`), the reserved `240.0.0.0/4` block (including the `255.255.255.255` broadcast address), and IPv6 site-local (`fec0::/10`) and multicast (`ff00::/8`). The validator now strips trailing dots before the hostname checks and fully expands IPv6 addresses to detect embedded private IPv4 targets. The download helpers now follow redirects manually (`redirect: 'manual'`), re-validating each hop before requesting it, so an unsafe redirect target is never fetched. When a redirect cannot be inspected because the runtime returns an opaque response, the helpers fail closed (reject the redirect) on the server; only in a real browser — where SSRF is not reachable (fetch is constrained by CORS and cannot reach a server's internal network or cloud-metadata endpoints) — is the redirect followed natively so legitimate redirected downloads keep working. - f18b08f: fix: redact server error details from UI message streams by default `toUIMessageStream`, `createUIMessageStream`, and `toUIMessageChunk` defaulted their `onError` callback to `getErrorMessage`, which serializes the raw error (`error.toString()` / `JSON.stringify(error)`) into the client-facing `{ type: 'error', errorText }` chunk — and also into `tool-output-error` parts. The documented default was `() => 'An error occurred.'`, so applications relying on the documented behavior were unknowingly streaming server exception details (internal hostnames, paths, provider request data, validation inputs) to end users. The default `onError` now returns the documented generic `'An error occurred.'`. Raw error details are only emitted when the developer explicitly supplies an `onError` handler. This also redacts `tool-output-error` and invalid-tool-input error text by default; pass an `onError` to surface richer messages. - b4507d5: fix(provider-utils): cancel response body on download rejection to prevent socket leak When a download was rejected early — because the `Content-Length` header exceeded the size limit, the response status was not ok, or a redirect resolved to a blocked URL — the fetch response body was left unconsumed and uncancelled. With WHATWG Fetch/undici this leaves the underlying TCP socket open instead of returning it to the connection pool, allowing an attacker-controlled origin to exhaust file descriptors and cause a denial of service. The body is now cancelled on all early-rejection paths in `readResponseWithSizeLimit`, `download`, and `downloadBlob`, and `fetchWithValidatedRedirects` cancels each redirect hop's body before following or rejecting the next hop. - Updated dependencies [8c17bf8] - Updated dependencies [aeda373] - Updated dependencies [558777f] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/gateway@4.0.0-canary.104 - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/alibaba@2.0.0-canary.59 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/amazon-bedrock@5.0.0-canary.84 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/anthropic@4.0.0-canary.66 - @ai-sdk/openai@4.0.0-canary.73 ## @ai-sdk/angular@3.0.0-canary.172 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/anthropic@4.0.0-canary.66 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/anthropic-aws@1.0.0-canary.7 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/anthropic@4.0.0-canary.66 ## @ai-sdk/assemblyai@3.0.0-canary.50 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/azure@4.0.0-canary.75 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/deepseek@3.0.0-canary.54 - @ai-sdk/openai@4.0.0-canary.73 ## @ai-sdk/baseten@2.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/black-forest-labs@2.0.0-canary.50 ### Patch Changes - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/bytedance@2.0.0-canary.51 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/cerebras@3.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/cohere@4.0.0-canary.52 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/deepgram@3.0.0-canary.49 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/deepinfra@3.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/deepseek@3.0.0-canary.54 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/devtools@1.0.0-canary.30 ### Patch Changes - 32ce45c: fix(devtools): secure devtools viewer api access - 275674e: fix(devtools): validate viewer database paths ## @ai-sdk/elevenlabs@3.0.0-canary.49 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/fal@3.0.0-canary.49 ### Patch Changes - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/fireworks@3.0.0-canary.56 ### Patch Changes - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/gateway@4.0.0-canary.104 ### Patch Changes - 8c17bf8: fix(gateway): surface provider warnings in embedding and reranking responses - 558777f: fix(gateway): accept deprecated warnings in image, speech, transcription, and video responses - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/gladia@3.0.0-canary.49 ### Patch Changes - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/google@4.0.0-canary.81 ### Patch Changes - 5878b40: fix(google): prevent prototype pollution when streaming tool args - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/google-vertex@5.0.0-canary.107 ### Patch Changes - Updated dependencies [5878b40] - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/google@4.0.0-canary.81 - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/anthropic@4.0.0-canary.66 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/groq@4.0.0-canary.53 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/harness@1.0.0-canary.7 ### Patch Changes - 3d87086: fix(harness): guard against invalid resuming a session vs continuing a turn - 1ea15a3: fix(harness): fix various bugs with harness skills not being correctly processed by the harness adapters - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/harness-claude-code@1.0.0-canary.3 ### Patch Changes - 1ea15a3: fix(harness): fix various bugs with harness skills not being correctly processed by the harness adapters - e551763: fix(harness): avoid using peer dependencies for underlying harness and sandbox SDKs - Updated dependencies [3d87086] - Updated dependencies [aeda373] - Updated dependencies [1ea15a3] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/harness@1.0.0-canary.7 - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/harness-codex@1.0.0-canary.3 ### Patch Changes - 1ea15a3: fix(harness): fix various bugs with harness skills not being correctly processed by the harness adapters - e551763: fix(harness): avoid using peer dependencies for underlying harness and sandbox SDKs - Updated dependencies [3d87086] - Updated dependencies [aeda373] - Updated dependencies [1ea15a3] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/harness@1.0.0-canary.7 - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/harness-pi@1.0.0-canary.3 ### Patch Changes - 1ea15a3: fix(harness): fix various bugs with harness skills not being correctly processed by the harness adapters - e551763: fix(harness): avoid using peer dependencies for underlying harness and sandbox SDKs - Updated dependencies [3d87086] - Updated dependencies [aeda373] - Updated dependencies [1ea15a3] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/harness@1.0.0-canary.7 - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/huggingface@2.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/hume@3.0.0-canary.49 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/klingai@4.0.0-canary.50 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/langchain@3.0.0-canary.172 ### Patch Changes - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - ai@7.0.0-canary.172 ## @ai-sdk/llamaindex@3.0.0-canary.172 ### Patch Changes - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - ai@7.0.0-canary.172 ## @ai-sdk/lmnt@3.0.0-canary.49 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/luma@3.0.0-canary.49 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/mcp@2.0.0-canary.63 ### Patch Changes - 2a150f8: fix(mcp): lock first sse endpoint received via event - b44b051: fix(mcp): prevent prototype-named tools from bypassing the `schemas` allowlist When using `client.tools({ schemas })` to expose only an explicitly allowed subset of an MCP server's tools, the allowlist check used the `in` operator, which also matches inherited `Object.prototype` properties. A server-advertised tool named `constructor`, `toString`, `__proto__`, etc. would pass the check even though the developer never defined it in `schemas`, and was then exposed to the model and executable. The check now uses `Object.hasOwn`, so only explicitly defined tools are returned. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/mistral@4.0.0-canary.54 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/moonshotai@3.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/open-responses@2.0.0-canary.53 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/openai@4.0.0-canary.73 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/openai-compatible@3.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/otel@1.0.0-canary.118 ### Patch Changes - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - ai@7.0.0-canary.172 ## @ai-sdk/perplexity@4.0.0-canary.52 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/policy-opa@1.0.0-canary.9 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/prodia@2.0.0-canary.52 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/provider-utils@5.0.0-canary.48 ### Patch Changes - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - 375fdd7: fix: harden download URL SSRF guard against hostname and redirect bypasses `validateDownloadUrl` and the file download helpers (`downloadBlob`, `download`) could be bypassed in several ways when handling untrusted URLs: - A fully-qualified hostname with a trailing dot (e.g. `localhost.`, `myhost.local.`) skipped the localhost/`.local` blocklist. - IPv6 addresses that embed an IPv4 address in their last 32 bits — IPv4-compatible (`::127.0.0.1`), IPv4-translated (`::ffff:0:127.0.0.1`), and NAT64 (`64:ff9b::127.0.0.1`, including the `64:ff9b:1::/48` local-use prefix) — were not decoded and checked against the private IPv4 ranges. - Redirects were validated only _after_ `fetch` had already followed them, so the request to a redirect target (e.g. an internal/metadata address) had already been issued before the check ran. - Several reserved/internal address ranges were not blocked: CGNAT (`100.64.0.0/10`, used by some cloud providers for internal traffic), benchmarking (`198.18.0.0/15`), IETF protocol assignments (`192.0.0.0/24`), the reserved `240.0.0.0/4` block (including the `255.255.255.255` broadcast address), and IPv6 site-local (`fec0::/10`) and multicast (`ff00::/8`). The validator now strips trailing dots before the hostname checks and fully expands IPv6 addresses to detect embedded private IPv4 targets. The download helpers now follow redirects manually (`redirect: 'manual'`), re-validating each hop before requesting it, so an unsafe redirect target is never fetched. When a redirect cannot be inspected because the runtime returns an opaque response, the helpers fail closed (reject the redirect) on the server; only in a real browser — where SSRF is not reachable (fetch is constrained by CORS and cannot reach a server's internal network or cloud-metadata endpoints) — is the redirect followed natively so legitimate redirected downloads keep working. - b4507d5: fix(provider-utils): cancel response body on download rejection to prevent socket leak When a download was rejected early — because the `Content-Length` header exceeded the size limit, the response status was not ok, or a redirect resolved to a blocked URL — the fetch response body was left unconsumed and uncancelled. With WHATWG Fetch/undici this leaves the underlying TCP socket open instead of returning it to the connection pool, allowing an attacker-controlled origin to exhaust file descriptors and cause a denial of service. The body is now cancelled on all early-rejection paths in `readResponseWithSizeLimit`, `download`, and `downloadBlob`, and `fetchWithValidatedRedirects` cancels each redirect hop's body before following or rejecting the next hop. ## @ai-sdk/quiverai@2.0.0-canary.5 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/react@4.0.0-canary.175 ### Patch Changes - 555c5de: fix(react): deny MCP App tool calls by default when allowedTools is omitted `experimental_MCPAppRenderer`'s bridge only enforced the `allowedTools` allowlist when it was non-null, so omitting `allowedTools` skipped the check and forwarded every `tools/call` from the (untrusted) MCP App iframe to the host's `callTool`. A malicious or compromised MCP server could therefore invoke any tool the host wired up. Tool invocation from MCP App content is now deny-by-default: if `allowedTools` is not explicitly provided, all `tools/call` requests are rejected. To expose tools to an app, list them in `handlers.allowedTools`. - Updated dependencies [2a150f8] - Updated dependencies [b44b051] - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/mcp@2.0.0-canary.63 - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/replicate@3.0.0-canary.50 ### Patch Changes - aeda373: fix: only send provider credentials to same-origin response-supplied URLs Several provider clients followed a URL taken from the provider's API response (a polling/status URL or a final media URL such as `polling_url`, `urls.get`, `result_url`, `result.sample`, or `video.uri`) and reused the authenticated headers — or appended `?key=<API_KEY>` — on that request. Because the host of the response-supplied URL was never validated, the long-lived API key was sent to whatever host the response named (a CDN in the benign case, or an attacker-chosen host if the provider response was tampered with), allowing credential exfiltration. A new `isSameOrigin` helper is added to `@ai-sdk/provider-utils`, and the affected fetches in `@ai-sdk/black-forest-labs`, `@ai-sdk/fireworks`, `@ai-sdk/replicate`, `@ai-sdk/gladia`, `@ai-sdk/fal`, and `@ai-sdk/google` now attach credentials only when the followed URL is same-origin with the provider's configured API origin. Requests to a foreign origin are made without the credential. - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/revai@3.0.0-canary.50 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/rsc@3.0.0-canary.173 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/sandbox-just-bash@1.0.0-canary.7 ### Patch Changes - e551763: fix(harness): avoid using peer dependencies for underlying harness and sandbox SDKs - Updated dependencies [3d87086] - Updated dependencies [aeda373] - Updated dependencies [1ea15a3] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/harness@1.0.0-canary.7 - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/sandbox-vercel@1.0.0-canary.7 ### Patch Changes - e551763: fix(harness): avoid using peer dependencies for underlying harness and sandbox SDKs - Updated dependencies [3d87086] - Updated dependencies [aeda373] - Updated dependencies [1ea15a3] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/harness@1.0.0-canary.7 - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/svelte@5.0.0-canary.172 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/togetherai@3.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/tui@1.0.0-canary.7 ### Patch Changes - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - ai@7.0.0-canary.172 ## @ai-sdk/valibot@3.0.0-canary.48 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/vercel@3.0.0-canary.56 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 ## @ai-sdk/voyage@2.0.0-canary.23 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 ## @ai-sdk/vue@4.0.0-canary.172 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/workflow@1.0.0-canary.89 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [25a64f8] - Updated dependencies [375fdd7] - Updated dependencies [f18b08f] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - ai@7.0.0-canary.172 ## @ai-sdk/xai@4.0.0-canary.74 ### Patch Changes - Updated dependencies [aeda373] - Updated dependencies [375fdd7] - Updated dependencies [b4507d5] - @ai-sdk/provider-utils@5.0.0-canary.48 - @ai-sdk/openai-compatible@3.0.0-canary.56 Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
PreviousNext