Verifiable Intent: Cryptographic Anchoring for Autonomous Agency

ABSTRACT

To counter the rise of Agent Hijacking, we propose a framework for 'Verifiable Intent.' This protocol cryptographically anchors the human user's original objective to every sub-task generated by a downstream agent. By using zero-knowledge proofs (ZKPs) to verify that an agent's proposed action is a direct and necessary descendant of the original intent, we create a 'Chain of Authority' that is resistant to redirection. We evaluate our framework on a multi-agent procurement system, demonstrating near-zero overhead while preventing 99.4% of intent-redirection attacks.