iProov has launched a hosted service that requires a verified person to approve high-risk actions an AI agent is trying to carry out, with the open-source agentgateway project and authorization platform Cerbos as its first integrations. The company’s chief technology officer, Dominic Forrest, is presenting the service at AgntCon North America in San Jose this week.
The service implements the Human Approval and Presence Specification (HAPS) that iProov released in September. It is designed to produce cryptographically verifiable proof that the right person approved a specific action, using iProov’s Dynamic and Express face verification technologies, and it issues an auditable, non-repudiable certificate for logging and later reference.
When a Cerbos policy requires human approval, the agent’s action is held and iProov’s hosted HAPS service is called. The service shows the named approver the exact action and its scope. If they approve, they complete an iProov genuine human presence check, and iProov issues a signed, single-use credential bound to that action before the action is released. If any detail changes, such as the amount or destination of a payment, the credential no longer matches and the action is blocked.
Agentgateway is an AI agent gateway developed by Solo.io, and Cerbos is iProov’s first policy-engine partner. “Cerbos controls and secures what decisions AI Agents can make in an organization, and under what conditions,” said Alex Olivier, Cerbos’ chief product officer and co-founder. “Our partnership with iProov introduces a new level of control, enabling organizations to require approval from the right human before a high-consequence action can proceed.”
The HAPS specification remains open, so any gateway, policy engine or relying party can implement HAPS-compliant approval backed by iProov or by other forms of proof. The specification, a partial reference implementation in Rust and test vectors are published on GitHub under the Apache-2.0 license.
“For organizations to realize the full potential of agentic AI, they need to task agents with increasingly consequential responsibilities. But some actions must remain subject to human control, approval and accountability,” said Andrew Bud, iProov’s founder and CEO. “By requiring a genuine human face as part of the approval process, we secure the process against agent impersonation. Because no matter how smart they become, agents will never have faces.”
Sources: iProov
–
By the ID Tech Editorial Team
