Look at what happened. Then ask why.
How incident evidence can make a risk discussion more useful.
Incident monitoring · Updated 2026-09-27
From abstract risk to context
The OECD’s incident-monitoring work aims to build evidence about AI-related incidents and hazards. Context matters: the same broad technology can be used in very different settings.
Read carefully
A reported case is a starting point for investigation, not proof of how often an event occurs. Ask what is established, what is alleged, and what remains unclear.
An incident-reading checklist
Our suggestion: record the affected people, the system’s role, the evidence available, the response, and the unresolved questions. Compare mechanisms across cases rather than treating every headline as an equivalent data point.
SOURCE & FURTHER READINGOECD · AI risks and incidents