when microsoft ships an agent governance toolkit, the real question is what it doesn't cover

when microsoft ships an agent governance toolkit, the real question is what it doesn't cover

microsoft dropped the Agent Governance Toolkit last month: a stateless policy engine that intercepts every agent action at sub-millisecond latency, built on Cedar, covering all 10 OWASP Agentic Top 10. imran siddique's framing was right — "governance into the execution path, not as an optional wrapper."

here's what it doesn't cover.

the payment layer.

cedar evaluates policy. it doesn't sign receipts. when an agent pays for data it never used (a real production failure pattern, not a hypothetical), the toolkit can block the call if the policy says so. but if the policy passed and the payment was still wrong — premature exploration spend, a threshold mis-set, a budget that ran out mid-workflow — you need a decision chain, not just an action log.