Reported AI harm cases rise to 346
Publicly reported AI failures and abuses are rising as AI moves into more places where mistakes can hurt people.
AI harm is no longer one story about one chatbot. It is becoming a pattern: fake images, fake voices, bad advice, broken systems, and real people left to sort out what happened.
The public incident count rose from 149 in 2023 to 346 in 2025.
- AI tools are being used in more daily products and decisions
- deepfake impersonation and fake content are easier to create
- public reporting catches more failures as attention rises
- organizations often deploy faster than they build safety and review processes
Sources · check usOpen
You should treat AI output like a powerful first draft, not proof. The more it enters daily life, the more your edge is knowing when to stop and verify.
Behind the numbersOpen
Observed values are from AI Incident Database data as reported through the Stanford AI Index and Our World in Data. The 2025 Stanford AI Index says reported AI-related incidents reached 233 in 2024, a 56.4% increase over 2023, which gives 149 for 2023. A 2026 reporting update cites 346 incidents for 2025. Our World in Data notes that the AI Incident Database relies on publicly available sources, entries are reviewed by human editors, incidents are likely undercounted, and past-year totals can change as the database updates. Projection is a cautious direction call because reporting intensity can change the count even if real-world risk changes differently.