Introducing Human Approval and Presence Specification: HAPS
Technical paper
An AI agent can arrive at a relying party with technically valid but agent-accessible evidence. This fails to establish that the human principal intended the specific action now being requested.
This iProov technical paper introduces the Human Approval and Presence Specification, an experimental procedural specification for binding human approval to an exact agentic action. It is published openly on GitHub under Apache-2.0 for industry review, critique, and independent implementation.
How does HAPS make human approval verifiable?
The specification defines four components:
- Action Intent: a canonical, machine-readable description of what the agent proposes to do.
- HAPS Challenge: optional and recommended for high-risk actions, allowing the relying party to state the proof of human presence and identity it requires.
- Signing View: the authoritative representation of the action presented to the human approver.
- HAPS Consent Credential: the resulting evidence, packaged so it can travel back to the relying party in portable, verifiable form.
What You’ll Learn
Follow the full sequence, from the relying party holding a pending action and issuing a challenge, through the presence check and explicit approval, to verification that the credential is genuine, sufficiently strong, current, and unused before the action is released. Understand how HAPS relates to OAuth, AuthZEN, OpenID4VP, and MCP, what the specification deliberately does not attempt, and how the accompanying Rust reference implementation and test kit support experimentation.
HAPS is proof-agnostic and implementation-neutral. The relying party retains responsibility for deciding which actions require approval, whether the approver is authorized, and how the action is held and executed.
View the specification on GitHub