Principles Applied
Never trust, always verify. Least privilege by default. Explicit verification per request. Continuous monitoring. These principles apply as much to AI agents as to users and services.
Agent Identity
Every agent has a verifiable identity. Per-agent credentials, not shared. Mutual TLS between agent and services where supported. No ‘agents are part of the platform so they’re trusted’ thinking.
Continuous Verification
Token refresh on short cadence. Behavior baselining — anomalous calls trigger re-auth. Conditional access (agent operating outside normal hours? Additional verification required).
Operational Reality
Zero-trust for agents adds friction. Balance security with usability. Customer-facing agents earn more friction than internal knowledge agents. Risk-weight your posture.