iProov has launched a new human approval service designed to strengthen security and accountability as enterprises deploy autonomous AI agents to perform high-risk tasks.
Based on the Human Approval and Presence Specification (HAPS), released in September 2026, the service provides cryptographically verifiable proof that the correct human has approved a specific action before an AI agent can execute it.
The service integrates with agentgateway, the open-source AI agent gateway developed by solo.io, and Cerbos, an authorization management platform, as its first policy-engine partner.
The HAPS engine combines iProov’s Dynamic and Express liveness technologies with verification that a person has approved a specific action. It then generates a signed, auditable certificate that organizations can retain for future reference.
How the human approval process works
When a Cerbos policy requires human approval, the AI agent’s requested action is placed on hold while iProov’s hosted HAPS service is called. The designated approver is shown the exact action and its scope before being asked to approve it.
The individual must then complete an iProov genuine human presence check. Once verified, iProov issues a signed, single-use credential tied to the approved action. Only then can the AI agent proceed.
The credential also prevents unauthorized changes. If details such as a payment amount or destination are modified after approval, the credential will no longer match, and the action will be blocked.
According to iProov, the system is designed to prevent AI agents from impersonating human approvers while maintaining a verifiable record of human authorization.
The announcement comes amid growing industry attention to security controls for autonomous AI agents. A Microsoft Security blog published in October 2026 highlighted the need for policies that can require cryptographically provable human approval before high-impact actions are carried out.
Cerbos determines when human approval is required and invokes the HAPS service to verify the designated approver. The integration is intended to let organizations automate more tasks through AI agents while retaining human control over consequential decisions.
The HAPS specification remains open, allowing other gateways, policy engines and relying parties to implement compliant approval workflows using iProov or other forms of proof. Its specification, a partial Rust reference implementation and test vectors are available on GitHub under the Apache-2.0 license.
iProov founder and CEO Andrew Bud said the service is intended to help organizations expand their use of agentic AI without compromising human oversight and accountability.
