Every AI agent safety incident provides information that, if captured correctly, prevents future incidents. A post-incident review (PIR) is the mechanism for extracting that information. This template ensures consistency and completeness across reviews.
Construct a minute-by-minute timeline from audit trail receipts and monitoring logs.
| Time | Event | Source | |------|-------|--------| | | First anomalous action detected | Sentinel alert | | | Incident confirmed by on-call engineer | Manual verification | | | Agent placed in shadow mode | Kill switch activated | | | Root cause identified | Audit trail analysis | | | Fix deployed | Policy update | | | Agent restored to full operation | Manual approval |
Answer these questions:
| Action | Owner | Due Date | Status | |--------|-------|----------|--------| | | | | |
Every action item must have an owner and a due date. Track completion in your project management system. Review open action items from previous PIRs at the start of each new review to ensure follow-through.
Explore more guides on AI agent safety, prompt injection, and building secure systems.
View All Guides