What are the steps of incident response for autonomous AI?
Detect, correlate, verify, classify, contain, revoke, preserve evidence, recover and review. Automated containment (session suspension, tool restriction, network isolation, credential revocation and forensic logging) buys time, and each action records its policy reference, timestamp, reason, affected identity, affected resource and audit record. High-impact decisions wait for human approval.
- Nine response steps.
- Six incident states: open, investigating, contained, isolated, recovering, resolved.
- Risk-based routing to humans.
Frequently asked questions
Which actions are automatic?
Low-risk actions run automatically, medium-risk ones are restricted or reviewed, high-impact actions need human approval, and critical or unauthorized actions are blocked and the agent isolated.