Agent Incident Registry logs 529 verified agent failures since 2022, separates harm from demos
anacondainc · x · 2026-09-23
The Agent Incident Registry (AIR) has tracked 529 verified AI agent incidents since 2022, separating realized harm from research demos. Every record requires a fetched source with verbatim quotes; unverifiable leads are quarantined. Incidents are coded into four classes (in-the-wild, safety failure, disclosed vulnerability, research demo), flagged for realized harm, and given permanent AIR-YYYY-NNNN ids with open exports — making agent failures citable, comparable, and learnable.
More from Safety
- Oxford's Yarin Gal Proposes arXiv Ban LLM-Written Papers to Curb AI Slop — yaringal · 2026-09-23
- Claude system card reveals METR's internal-access team shared conclusions, not evidence — rohanpaul_ai · 2026-09-23
- $1B and unlimited frontier tokens: where would you spend them to fix cybersecurity? — chrisrohlf · 2026-09-23
- Stanford accused of using AI to alter students' race, gender and body in ads — soleio · 2026-09-23
- Grady Booch refuses to install Meta's Muse agent, citing distrust of Meta — Grady_Booch · 2026-09-23
- Anthropic's Claude Opus 5.5 system card adopts external evaluation-awareness framework — maksym_andr · 2026-09-23