AI incident response: write the plan before the incident

In short: an AI incident is any event in which an AI system causes or nearly causes harm, breaks policy or law, or behaves outside its approved design. Some are now legally reportable on short clocks, from 24 hours to 15 days. Every system that matters needs a one-page procedure: severity levels, who can pause it, who is told, what evidence is kept, how people are informed, and how the lesson returns to testing.^1,2,3

Stanford’s 2026 AI Index counted 362 documented AI incidents in 2025, up from 233 the year before.¹ As AI moves into customer service, hiring, finance and operations, an incident is no longer a hypothetical for most organizations. The ones that handle it well have one thing in common: a process written before it was needed.