Skip to content

2026.11.2

Latest

Choose a tag to compare

@MarouaBoud MarouaBoud released this 18 Nov 05:02

A versioned snapshot of the agent-postmortems corpus — a structured, public database of real AI-agent failures.

This snapshot

  • 180 incidents (101 realized · 79 demonstrated hazards / near-misses), 2023–2026
  • 18 of 19 failure classes represented; 40 CVE-backed
  • Every record conforms to the v1 schema and is classified against the versioned taxonomy

Since 2026.11.1

  • 32 new records (10 realized · 22 hazards) from the 2025–2026 historical backlog
  • Realized: the Wiz CIRT agent-run multi-cloud intrusion, Symantec's "Skynet Mode" LLMjacking agent, Microsoft's LiteLLM, RAGFlow and Kestra control-plane intrusions, the CISA KEV listings for Langflow (CVE-2026-9198) and for LiteLLM and Kestra, Barracuda's inbox-assistant phishing injections, Tencent's evaluation agent fleet, the BMW dealership chatbot offer, Meta Muse's address disclosure, and OpenAI's third-party notifications following its model-activity review
  • Hazards: seven Embrace The Red demonstrations (Amp Code and Cline image exfiltration, hidden Unicode in MCP tool metadata and in agent skills, ChatGPT deep-research connector spill, AgentHopper, Claude Code auto-mode classifier bypass), Ghostcommit, Grok cryptographic context injection, Workflow Identity Hijacking, Datadog's counterfeit-policy classifier bypass, the Cathedral test against Anthropic, MONST3R-AI, and nine framework advisories with no reported exploitation (Langflow CSV agent, Neuron, Cursor CLI, FastGPT, Daytona, Pydantic AI, DB-GPT, Vibe-Trading, Loom for AWS)

Assets

  • incidents.json — the full corpus as one JSON array
  • scenarios.json — replayable scenario shapes (for stampede)
  • seeds.json — denial-of-wallet seeds (for costbomb)

Export entries carry natural-language trigger shapes, never runnable exploit payloads.

Use the data

Incident data is licensed CC-BY-4.0 — reuse freely with attribution. Cite via CITATION.cff.