Who certifies AI agents today?
The accounting profession moved first. In October 2026 the AICPA, the US body behind SOC 2, announced a program with the startup AIUC pairing SOC 2 attestation with AIUC-1, presented as the first comprehensive independent certification program for AI agents (CFO Dive on the AICPA and AIUC program, read 26 October 2026). The tested failure modes are agent-specific, among them prompt injection, where instructions hidden in an email or document manipulate the agent. Certification extends the audit profession’s trust machinery to a new subject, the same move SOC 2 once made for cloud software, and it sits inside the broader market for AI assurance.
Why would a buyer ask for it?
Because vendor claims about agent safety are unverifiable from outside. An agent that reads documents and takes actions has failure modes a demo never shows, and procurement teams lack the means to red-team every vendor. A certificate moves part of that burden to an independent assessor with a published standard, the role ISO 42001 plays for AI management systems and frameworks like AI TRiSM play inside the enterprise. The certificate’s value is exactly the independence and the standard behind it, so the first question about any badge is who issued it and against what.
What should a buyer still do themselves?
Everything that depends on their context. A certification tests the agent as shipped, against general standards, at a point in time; it does not know your data, your permissions, or the model update the vendor ships next quarter. Keep your own agent governance: scoped access, approval gates on consequential actions, logging, and an owner per agent. Ask how the certificate handles change, what triggers re-assessment, and what the vendor warrants contractually when the certified behavior fails. Treat certification as evidence in procurement, not as a transfer of responsibility; the people-credential market next door, AI certification, teaches the same lesson about badges.