AIAcademy · AIAcademy · 2026-05-16
AI Incident Database — recent court-filing incidents
1,227 catalogued court filings with AI-hallucinated citations as of May 2026, per the running tally in the AI Incident Database. A $110,000 sanction issued by an Oregon district court in April 2026 — the largest hallucination-related fine yet. The first license suspension in Nebraska. Sixth Circuit and Third Circuit precedent now explicitly holding that Rule 11 obligations attach to AI-generated content. The AI Lawsuit Tracker keeps the running docket.
If you bill hours and use a model, this is now a real career risk. The checklist below is the minimum any professional handling legal, medical, or financial work should be able to defend in front of a regulator.
1. Verify every citation against a primary source. Not against another AI summary. Not against a "summary of relevant case law" that the same model produced. Open the actual reporter, the actual journal, the actual filing. Hallucinations cluster in citation structure — plausible-looking case names with wrong volume numbers, real authors with invented papers.
2. Disclose AI use when the forum requires it. Standing orders in dozens of federal districts now mandate disclosure. The recent incident summaries include several sanctions where the underlying error was minor but the failure-to-disclose was dispositive.
3. Keep the prompt and the output. A reviewable record of what you asked and what the model returned is the difference between a correctable mistake and a fitness-to-practice complaint. Treat the chat log like a working file.
4. Never let the model do the final read. A human professional signs off. The malpractice insurance, the bar admission, the medical license, the fiduciary duty — none of those run to the model.