Documented artificial intelligence incidents reached a record level in 2025, according to data compiled for Stanford University's 2026 AI Index, underscoring how the risks tied to AI deployment are growing alongside adoption. The AI Incident Database recorded 362 documented incidents in 2025, up from 233 in 2024, a 55 percent increase year over year.

The database defines an incident as a harm or near harm realized in the real world through the deployment of an AI system, a category that spans data exposures, safety failures, discriminatory outcomes, and other documented harms. The 2025 total continues a steep upward trend, following a 56.4 percent jump the prior year that itself set a record.

The report pairs the rise in incidents with a decline in transparency. The Foundation Model Transparency Index fell from an average score of 58 to 40, indicating that developers are disclosing less about how their systems are built and governed even as those systems spread into more products and workflows.

The combination presents a widening governance gap. Adoption is climbing, incidents are climbing, and transparency is falling at the same time, a pattern researchers describe as making oversight harder rather than easier. The figures reinforce why data protection, model documentation, and incident tracking have become central concerns for organizations deploying AI at scale.

Source: Stanford HAI - https://hai.stanford.edu/ai-index/2026-ai-index-report/responsible-ai