8000 GitHub - agnivesh/ai-hack-watch: A public timeline of notable AI hacking and AI-enabled cyber incidents. · GitHub
[go: up one dir, main page]

Skip to content

About

A public timeline of notable AI hacking and AI-enabled cyber incidents.

Topics

Resources

Contributing

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

AI Hack Watch

A static single-page site tracking notable AI hacking and AI-enabled cyber incidents.

Days since last AI hack

Features

  • Large "days since" counter.
  • Chronological public timeline.
  • Structured metadata for AI role and incident category.
  • Filters by AI role/category.
  • Click any role or category label (filter bar, story tags, or breakdown) to show only matching incidents.
  • Incident counts and longest observed streak.
  • RSS feed generated automatically.
  • GitHub Issue Form for non-technical story submissions.
  • Pull request template and contribution guide.
  • Automated PR validation.
  • Automated pipeline tests (scripts/test_pipeline.py).
  • GitHub Action that turns data.md into incidents.json, feed.xml, llms.txt, and a pre-rendered recent timeline.

Data workflow

You edit data.md.

data.md
   ↓
Pull request / merge
   ↓
GitHub Action
   ↓
incidents.json + feed.xml + index.html
   ↓
GitHub Pages

The generated files and the pre-rendered timeline should not normally be edited manually.

Article format

---

## Article headline

**Date:** 2026-09-19
**Source:** Example Security Research
**URL:** https://example.com/article
**AI role:** Autonomous
**Category:** Security research

Short factual description of the story.

Required metadata

  • Date — ISO 8601 calendar date: YYYY-MM-DD
  • Source
  • URL
  • AI role
  • Category

Multiple sources

If a story has more than one authoritative source, add optional numbered fields so readers still have a working link if one publisher goes down:

**Source:** Sysdig Threat Research Team
**URL:** https://www.sysdig.com/blog/jadepuffer-agentic-ransomware-for-automated-database-extortion
**Source 2:** Dark Reading
**URL 2:** https://www.darkreading.com/cyberattacks-data-breaches/jadepuffer-first-complete-llm-driven-ransomware-attack

All source URLs appear on the timeline; the first is the primary citation.

AI roles

  • Autonomous
  • AI-assisted
  • AI-targeted
  • AI security research
  • Unclear

Ordering

Contributors do not need to insert an article in the right chronological position. The Action sorts generated data by date, newest first.

Contributions

Read CONTRIBUTING.md.

You can contribute in two ways:

  1. Edit data.md in a pull request.
  2. Use the Submit a story GitHub Issue Form.

The site links visitors to the issue form. The repository URL is set in config.js:

window.SITE_CONFIG = {
  repositoryUrl: "https://github.com/agnivesh/ai-hack-watch",
  siteUrl: "https://aihackwatch.com"
};

Methodology

Read METHODOLOGY.md for what qualifies, AI-role definitions, source preferences, date rules, corrections, and calculation details.

Development

Run the pipeline and its tests locally (skipping Wayback archiving, which would hit the network):

python3 scripts/test_pipeline.py          # regression tests
python3 scripts/validate_contribution.py  # same checks as PR CI
python3 scripts/generate.py --no-archive  # regenerate incidents.json, badge, feed.xml, llms.txt
python3 scripts/prerender_timeline.py     # regenerate the in-page timeline + OG tags
python3 scripts/check_links.py            # HEAD-check source URLs

generate.py resolves missing Wayback links in-memory by default without rewriting data.md. Pass --write-archives only on the push path (CI does this automatically; daily badge refreshes use --no-archive) to backfill Archive: fields.

The pipeline lives in scripts/ (generate.py, validate_contribution.py, prerender_timeline.py, check_links.py) and shares one parser/validator in datamd.py.

GitHub Pages

The site is deployed to https://aihackwatch.com (replacing the default https://agnivesh.github.io/ai-hack-watch/ URL). The CNAME file at the repo root declares the custom domain; DNS points the domain at GitHub Pages and HTTPS is provisioned automatically.

To deploy your own copy of the project:

  1. Create a GitHub repository.
  2. Upload the contents of this project.
  3. Update config.js with your repository URL.
  4. In Settings → Pages, choose Deploy from a branch.
  5. Select the main branch and / (root).
  6. Save.

Actions permissions

The generation workflow commits incidents.json, feed.xml, llms.txt, and the pre-rendered timeline in index.html back to the repository.

In Settings → Actions → General, make sure workflow permissions allow repository contents to be written. The workflow also declares:

permissions:
  contents: write

RSS

feed.xml is regenerated whenever data.md changes.

Search visibility

The generation workflow embeds the 20 most recent incidents directly in index.html. Search engines can read these stories without executing JavaScript, while visitors still receive the complete filterable timeline from incidents.json.

Validation

Pull requests run .github/workflows/validate-contribution.yml, which checks:

  • Required metadata.
  • ISO date format.
  • Future dates.
  • Valid HTTP(S) URLs.
  • Valid AI role values.
  • Archive URL format (when provided).
  • Duplicate URLs.

Pull requests and pushes also run .github/workflows/test.yml, which executes the regression suite in scripts/test_pipeline.py. The validation workflow is intentionally conservative; maintainers still review factual relevance and source quality.

Embeddable badge

![Days since last AI hack](https://aihackwatch.com/days-since-badge.svg)

The badge is regenerated daily (and on every data change). Opened directly in a browser, it also updates itself live from the latest tracked date.

Machine-readable feeds

  • incidents.json — public JSON dataset. Contains schema_version, chunk_size, total, and incidents array. See Schema versioning.
  • feed.xml — RSS feed.
  • llms.txt — site overview for LLMs and agents, following llmstxt.org.

Schema versioning

incidents.json includes a schema_version field (currently 3) to allow future format evolution without breaking consumers. Version 2 added sources and archives, both index-aligned with urls so each source can carry its own Wayback snapshot; the top-level archive field remains the primary snapshot for older consumers. Version 3 adds updated, the date of the latest material revision, while date stays the first public disclosure. The chunk_size field (currently 20) hints at a future pagination strategy if the dataset grows large.

Dataset size guideline: incidents.json should remain under 250 KB (~1,500 incidents) to keep initial load fast on mobile. At the current rate (~20 incidents/year), this provides decades of headroom. If the dataset approaches this limit, maintainers should implement chunked pagination using the chunk_size hint.

Broken links

A weekly workflow checks source URLs and opens or updates a GitHub issue when a link appears unavailable. Automated link checks can produce false positives when publishers block bots, so the issue is for human review rather than automatic deletion.

Permalinks and sharing

Each incident has a slug-based #anchor so individual stories can be shared directly. The page also updates its title and Open Graph description based on the latest incident.

Automatic link archiving

When an incident has no Archive value, the GitHub Action looks for an existing Wayback snapshot before asking for a new one. Additional sources use numbered archives (Archive 2, Archive 3, ...), each filled independently, so a missing snapshot for one source no longer masks the others. Existing archive URLs are preserved. The timeline shows an Archived copy ↗ link for the primary source and an Archived ↗ link for each secondary source with a snapshot.

The check is availability-first because a save costs the service much more than a lookup. If the archive service reports two consecutive transient failures (rate limiting or an outage), the run stops asking and leaves the fields blank for the next run. That keeps a degraded Wayback Machine from receiving one request per unarchived URL every day.

Archiving is best-effort and never blocks a valid contribution.

About

A public timeline of notable AI hacking and AI-enabled cyber incidents.

Topics

Resources

Contributing

Stars

0 stars

Watchers

0 watching

Forks

Contributors

Languages

0