Who Verifies the Verifier? AI Agents Line Up for Identities

Credentials, mandates, and the edge

Indicio's proposal extends the identity architecture built for humans to the agents acting for them. Agents would carry signed verifiable credentials rather than keys, checked against machine-readable trust lists, with delegated authority encoded in the credential itself: what roles this agent holds, what it may do, and for whom. The design keeps authority at the edge, with the user's device holding keys and credentials while agents request signed mandates for specific actions instead of receiving standing API access.

The regulatory ground is moving in the same direction. In early September, FinCEN and four fellow regulators issued joint FAQs confirming banks may accept verifiable digital credentials in Customer Identification Program compliance, and the payments networks are already racing to define agent rails with Google's AP2, Visa's Trusted Agent Protocol, and Mastercard Agent Pay.

The verification gap in the middle

What the whitepaper concedes is as instructive as what it proposes. Indicio has demonstrated an agent verifying a human's credential, biometrics included, without human review. The larger vision, unfamiliar agents establishing mutual authentication and transacting at machine speed under delegated authority, remains a gap between demonstration and deployment, and that gap is precisely where Know Your Agent programs live.

Someone has to verify the human principal behind the agent at enrollment, bind the mandate to that verified identity, and re-verify when the mandate escalates, which is the role Arc plays in the deepidv stack: a gateway where agents present credentials, mandates trace to liveness-verified humans, and every delegation lands in an evidence trail. The API key era of agent identity is ending the way the password era did: not because something better was invented, but because the incident rate made the status quo unpriceable.