Definition
The controls that give each AI agent its own verifiable identity and grant only the short-lived, task-specific permissions it needs. NIST's design tenets call for unique cryptographic identities, zero standing privileges, just-in-time revocable access, and human review when an agent crosses a risk threshold.
Why it matters
Without it, an agent can inherit a person's broad access or be impersonated, so one compromised agent could approve its own actions and move sideways into sensitive systems.