iProov, a provider of biometric identity verification solutions, has released the Approval and Presence Specification (HAPS), aimed at enhancing AI governance and oversight. Designed to enable organizations to ensure human approval before AI agents execute specific actions, the specification responds to a growing need for stringent management of AI capabilities.
The HAPS framework, now available on GitHub under the Apache-2.0 license, comprises a series of rules targeting the governance of AI agents. As AI entities increasingly operate with greater autonomy, HAPS addresses the necessity for human oversight in scenarios where AI can inadvertently misuse delegated credentials through actions like prompt injection or excessive goal-setting, despite staying within permitted boundaries.
Tackling AI Challenges
The potential for AI agents to take unintended actions without explicit human consent highlights the need for a protocol like HAPS. It ensures that actions undertaken by AI have verifiable human approval, particularly in critical scenarios, providing a prototype solution for organizations to check human intent against what the agent intends to execute.
The HAPS initiative does not necessitate human approval for every AI action
Andrew Bud, founder and CEO of iProov, emphasizes the shift in AI capabilities: “AI agents are moving rapidly from answering questions to taking actions on our behalf. As their autonomy and capabilities grow, governance must keep pace.” He further explained, “We need to distinguish between an agent having permission to act and a human actually approving the specific action it is about to take.” Bud encourages industry-wide collaboration to scrutinize and enhance the specification, reinforcing the importance of community efforts to establish effective safeguards.
Focused Human Approval
The HAPS initiative does not necessitate human approval for every AI action, which could lead to approval fatigue. Instead, it empowers organizations to determine which actions require additional scrutiny based on their sensitivity or critical nature. HAPS facilitates a secure linkage between a human’s endorsement and the action in question, offering a means of verifying alignment between intended and proposed activities by AI agents.
For those interested in exploring the HAPS specification, reference implementation, and accompanying test materials, they are currently accessible on GitHub, marking a step forward in tackling the challenges of a rapidly evolving AI environment.
