Least privilege for AI agents: Identity, access, and tool binding

This guidance describes the security risks of deploying autonomous AI agents without proper identity and authorization controls and provides practical recommendations: treat each agent as a first-class principal with a dedicated lifecycle-managed identity, apply task-scoped least-privilege RBAC, enforce controlled tool access and allowlists, use just-in-time elevation for high-impact actions, and implement comprehensive end-to-end auditing and revocation testing to preserve accountability and reduce unintended impact.