Show HN: Arc – Delegated authority and audit for AI agent actions (github.com)

ARC (Authority and Audit for AI Agents) is a newly introduced protocol designed to provide a structure for granting, narrowing, approving, revoking, contesting, and adjudicating authority among AI agents and delegated systems. It emerged from challenges faced in agentic commerce, where questions of authority and trust in transactions predominated. ARC's focus is on ensuring that authority can be reliably delegated and audited across various implementations without dependence on a central trusted database, making it applicable to multiple domains beyond commerce. This protocol establishes five essential questions to clarify delegation and audit processes: who holds authority, the exact scope of that authority, what approvals cover actions, how revocation can occur, and how to recompute authority from experiences. By employing a compact Event/Projection model, ARC allows for the creation of portable authorization records that can traverse systems and be verified independently. Its potential impacts are significant for the AI/ML community, as the shared semantics of authority provided by ARC promise to streamline auditing and administrative interactions involving AI, fostering trust, and enhancing interoperability across different systems.