Reported AI harm incidents keep rising
Reported AI harms and near-harms hit a new high in 2025, showing the damage is no longer theoretical.
The scary part is not that AI sometimes fails. The scary part is that the failures now show up in elections, classrooms, courts, workplaces, scams, and private lives.
Reported AI incidents rose from 233 in 2024 to 346 in 2025.
- More people are using AI, so more failures reach the real world.
- Fake images, fake voices, and fake text make deception cheaper.
- Institutions are still building the habits needed to check AI before harm spreads.
Sources · check usOpen
You now live in a world where seeing is weaker proof than it used to be. A voice, photo, video, résumé, review, or answer may need a second layer of trust.
Behind the numbersOpen
The AI Incident Database collects public reports of cases where AI systems caused or nearly caused harm. Our World in Data republishes the annual count using the AI Incident Database via the Stanford AI Index. Stanford’s 2025 AI Index reported 233 incidents in 2024, a 56.4% increase over 2023, which implies about 149 incidents in 2023. The 2026 AI Index data series reports through 2025, and outside summaries of that dataset report 346 incidents in 2025. Caveat: this is a reported-incident count, not a full count of every harm. Media attention, reporting standards, and database updates can change the numbers.