Most AI governance writing this year has been about frameworks — principles, risk maps, voluntary standards. Less attention has gone to who actually checks whether an organization's agent deployment matches its paperwork. The Cloud Security Alliance's research arm flagged that gap directly in an April note on the AI agent governance shortfall, and it points to a concrete fix already underway rather than a proposal. The CSA's existing STAR program — its longstanding third-party assurance mechanism for cloud security claims — is being extended specifically for agentic AI. The research note describes the CSA STAR program's assurance mechanisms and the extended assurance capabilities planned under the new CSAI Foundation, announced in March 2026, as work that will give organizations a structured pathway for demonstrating agent governance to customers, partners, and regulators as this space matures. That's a narrower claim than it might sound, and worth being precise about. STAR was built for cloud providers making security claims that customers couldn't independently verify — a vendor says it does X, and a third party checks the paperwork and evidence behind that claim. Applying the same model to agents means an enterprise or vendor could eventually get an external attestation that its agent fleet has real identity controls, logging, and human-oversight checkpoints in place, not just a policy document saying it should. The CSA note pairs this with an existing technical checklist: it recommends organizations apply the MAESTRO threat-modeling framework together with the OWASP Agentic Top 10 for comprehensive threat coverage when doing agent security assessments. That's the diagnostic layer — MAESTRO and OWASP tell you what could go wrong. STAR-for-agents is meant to be the verification layer — an independent check that you actually addressed it, expressed in a form a third party (a customer's procurement team, a regulator, an auditor) can rely on without doing the work themselves. The distinction matters because most of what's shipped in agent governance so far is self-declared: vendors publish their own trust centers, their own security whitepapers, their own claims about what their agents do and don't do. An assurance program run by a body separate from the vendor is a different kind of evidence — closer to a SOC 2 report than a marketing page. Whether the CSAI Foundation's version gets real adoption depends on whether enough buyers start asking for the certificate instead of the whitepaper. That's not resolved yet, but the infrastructure to make the ask meaningful is now being built.