Stanford University's 2026 AI Index documents a steady rise in reported AI harms, reinforcing data from incident trackers that show the number of public AI failures climbing each year. The report frames the increase against a backdrop of rapid deployment, where organizations are adopting AI faster than they are building safeguards around it.

The pattern echoes figures from independent databases. Tracked AI incidents reached 362 in 2025, up from 233 the prior year, while reports rose roughly 50 percent year over year across the preceding period. The Index situates these counts within a wider analysis of responsible AI practices, noting that measurement and standardization of AI harms remain inconsistent across organizations and jurisdictions.

A central theme is the gap between capability and control. As models grow more capable and reach more users, the surface area for failures, from data exposures to harmful generated content, expands accordingly. The report highlights that inaccuracy has become the top-cited AI risk among organizations, with the share naming it rising sharply in a single year, a concern that sits directly upstream of many documented incidents.

The data points to a market in which AI use is widespread but the infrastructure for catching and reporting failures is still maturing. The Index suggests that as adoption deepens, the volume of incidents is likely to keep rising unless investment in evaluation, monitoring, and governance keeps pace. For organizations deploying AI, the figures underscore the value of testing and oversight before systems reach customers or the public.

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