Documented AI Failures

Largest pandemic fraud in U.S. history. Fake meal rosters used names from ‘listofrandomnames.com’ with Excel-generated ages. One restaurant claiming 5,000 meals/day previously served ~40 customers. Only 3% of funds went to food.

14 state-run Medicaid programs flagged for massive fraud. Housing program costs ballooned from $2.6M to $100M+. ‘Fraud tourism’ — people flew from Philadelphia to bill fake claims remotely.

Integrated Community Supports program exploded from $4.6M (2021) to $170M (2024). Providers billed for 12 hours of daily care while checking on clients once a day. One man died in 'care.'

Providers hired unqualified ‘behavioral technicians,’ submitted false claims, and paid parents $1,500 kickbacks to enroll children.

AI security system at a school mistook a student’s bag of Doritos for a firearm, triggering a full armed police response.

Two U.S. District Court judges forced to withdraw published rulings after discovering law clerks used ChatGPT and Perplexity, inserting factual errors into final judicial decisions.

Government Accountability Office began dismissing bid protests as sanctions when federal contractors submitted AI-generated legal filings containing hallucinations.

Multiple state attorneys general formally demanded Microsoft, OpenAI, and Google fix ‘delusional’ AI outputs that harm consumers and businesses.

First comprehensive U.S. state AI law takes effect requiring impact assessments, transparency disclosures, and documentation for high-risk AI systems.

Craig Guildford, Chief Constable of West Midlands Police (England’s third-largest force), resigned January 16, 2026 after Microsoft Copilot hallucinated a nonexistent football match in an official intelligence report used to ban Israeli fans from an Aston Villa Europa League game. Guildford denied to Parliament — twice — that any AI was involved, then admitted it in a letter to the Home Affairs Committee. The UK Home Secretary and Downing Street both publicly stated they had ‘lost confidence’ in him.