Proofs of Autonomy: Scalable and Practical Verification of AI Autonomy

Artem Grigor, Christian Schroeder de Witt, Ivan Martinovic

Keywords: Agent Proofs, Autonomous Agents, Verifiable Autonomy, Auditable AI, Agent Identity

Abstract: As AI agents increasingly manage critical financial, data, and legal processes autonomously, stakeholders require robust verification of agent-generated actions. We introduce \emph{Proofs of Autonomy}, a cryptographic framework that binds each agent action to a verifiable execution trace defined by an immutable \emph{Agent Card}. The Agent Card precisely specifies the agent’s configuration, capabilities, and accepted proof methods, enabling formal guarantees of completeness, soundness, non-repudiation, and privacy. We instantiate our framework practically using MPC-assisted TLS transcripts (\emph{Web Proofs}), enabling secure and auditable verification without modifications to proprietary models or external APIs. Our evaluation with a fully autonomous crypto-trading agent shows minimal verification overhead (under 2 seconds per interaction after initial setup), demonstrating that Proofs of Autonomy provide a scalable, efficient, and practical foundation for trustworthy AI autonomy.

Submission Number: 27