# Revocation & containment

Invalidate a compromised agent everywhere at once and contain the incident in seconds, on the record.

## One-action revocation <!-- #one-action-revocation -->

Revoking an identity invalidates its credentials and scopes across every connected system at once. There is no per-system rotation to chase — containment is a single call.

In practice the path from anomaly to full revocation is measured in seconds, not the hours it takes to rotate shared keys.

## When to revoke <!-- #when-to-revoke -->

Revoke on any signal that an agent is compromised or misbehaving: an out-of-scope request, an anomalous action rate, or a sponsor's call. Revocation is cheap and reversible in effect — re-register or re-issue when the situation is understood.

Because credentials are short-lived, revocation often just brings forward an expiry that was minutes away regardless.

## What gets recorded <!-- #what-gets-recorded -->

Every revocation writes to the audit log with the trigger, the actor, the timestamp, and the scopes withdrawn. The record is attestation-ready, so an incident review starts from facts, not reconstruction.

## After containment <!-- #after-containment -->

Once contained, inspect the agent's recent actions in the audit log, tighten policy or scope if needed, and issue fresh credentials to bring it back. Nothing standing survives the incident.
