[xizhuomengcontin]

Summary

Add the Orca AI Incident Archive to Governance, Safety & Responsible AI → Security & red-teaming, between the NIST taxonomy and OWASP so the cell stays alphabetical.

That cell has the frameworks (MITRE ATLAS, NIST AI 100-2, OWASP) and Simon Willison's prompt-injection series. The archive adds the case record behind them: 365 publicly disclosed agent security events from January 2025 on, one Markdown file each with at least one primary source and an attack-chain diagram. Each record has a real_harm field that separates a confirmed victim (133 records) from a researcher showing an attack could work. Examples include Replit Agent deleting a production database, the poisoned Amazon Q extension and Claude Code running terraform destroy on DataTalks.Club's production.

Resource metadata (for additions)

• Title: Orca AI Incident Archive
• URL: https://github.com/Continuum-AI-Corp/Orca-AI-Incident-Archive
• Author / org: Continuum AI (Continuum-AI-Corp)
• Resource-type row: Governance, Safety & Responsible AI
• Topic cell (T1–T21): Security & red-teaming (T15)
• Provenance (who cites it?): None yet, to be honest. It is new (created 2026-09-16) and has few stars. Severity and classification are the editors' judgement, not official findings. Publicly disclosed events only, no exploit code, CC BY 4.0. Disclosure: I'm on the team that maintains it. If you'd rather keep this cell to established references, I understand.

Checklist

• I read CONTRIBUTING.md.
• This is a resource (article/book/course/paper/talk/cookbook/benchmark/etc.), not a pure tool.
• I placed the entry in the strongest-fit matrix cell (no cross-posting except Trending/Milestones/Core).
• The link resolves and uses the canonical URL (no tracking params, no shorteners).
• The one-line description explains what you'll learn / why it matters, not marketing copy.
• The entry isn't already listed elsewhere in the README.
• Formatting matches neighbouring entries (order, icons, shields).
• awesome-lint and markdownlint pass locally (or CI will verify).

🤖 Generated with Claude Code