California attorney building open infrastructure for AI agents. Every project here is open source under Apache-2.0, and every capability claim traces to recorded evidence rather than a release note.
- Sanctuary: security for AI agents. An egress wall the operating system enforces (proven on macOS; proven on Ubuntu 24.04 x86-64 through a separate Castle Wall Linux package; nothing on Windows yet), keys only the operator holds, and a hash-chained record of what the agent did. Runs as an MCP server:
npx -y @sanctuary-framework/mcp-server@1.8.4. Site: sanctuaryprotocol.ai. Evidence: Assurance Matrix. - Concordia: an open standard for structured negotiation between agents, with signed multi-term offers, binding commitments and session receipts. Python:
pip install "concordia-protocol[server]". Site: concordiaprotocol.dev. - Verascore: reputation for agents, scored from signed records rather than self-reported claims.
- Claude Code plugins: install Sanctuary and Concordia from one marketplace.
Sanctuary and Concordia work alone and compose when both run; neither depends on the other.
- Co-chair, W3C Agentic Integrity Verification Specification (AIVS) Community Group
- X: @newtonlaw
- LinkedIn: erikwnewton