Trackers · Living reference

AI Incident & Failure Tracker

What actually went wrong — documented production AI failures, liability rulings, and court sanctions, each with a primary source. The reference for anyone writing an AI policy who needs precedent, not vibes. Updated as cases land.

Last updated 2026-06-20 · 3 incidents

DateSubjectWhat happenedAI involvedTypeSource
Oct 2025California attorney (Mostafavi)Fined $10,000 — the largest California AI-fabrication penalty to date — after 21 of 23 case quotes in an appellate brief turned out to be fabricated by ChatGPT.OpenAI — ChatGPTCourt sanctionThe Daily Recordthedailyrecord.com
Feb 2024Air CanadaHeld liable for negligent misrepresentation by its support chatbot (Moffatt v. Air Canada); the tribunal rejected the airline’s claim that the chatbot was a "separate legal entity".In-house support chatbotLiability rulingAmerican Bar Associationamericanbar.org
Jun 2023Mata v. Avianca (two NY attorneys)The landmark case: lawyers sanctioned after submitting a brief full of non-existent cases hallucinated by ChatGPT — the first widely-cited AI legal-citation failure.OpenAI — ChatGPTCourt sanctionSeyfarth Shawseyfarth.com

How this is compiled

  • Documented incidents only. Each row cites court/tribunal coverage, law-firm analysis, or Tier-1 reporting — not rumor or speculation.
  • Practitioner focus. We track the failures that set precedent for enterprise AI policy: who was held liable, who was sanctioned, and why.
  • Know an incident we’re missing? Tell us.

Informational only · not legal advice