Mysten Labs and Google Cloud Introduce AI Agent Evidence Layer on Sui

Mysten Labs and Google Cloud have introduced the Verifiable Agent Arbiter (VAA), an evidence layer designed to track and verify AI agent activity. The system generates cryptographic proofs of AI interactions, storing detailed records in customer-controlled Google Cloud Storage while keeping proofs on Walrus and coordinating them on the Sui blockchain. This setup aims to provide independently verifiable records of prompts, model outputs, tool calls, and policy decisions, addressing the need for reliable auditing and compliance with regulations like the EU AI Act.

The VAA separates detailed telemetry from cryptographic proofs, ensuring that security teams can reconstruct agent workflows to identify issues such as prompt-injection attacks, policy bypasses, or model failures. The system also supports replaying disputed workflows under the Agent2Agent (A2A) protocol, allowing parties to verify transactions and establish accountability. Additionally, the evidence can be re-proven under future cryptographic standards, including post-quantum methods.

VAA integrates with the Sui Agent Payments facilitator, enabling agents to discover services and settle payments for API calls under the x402 protocol. The system records authorization, spending decisions, and service delivery, enhancing transparency in AI agent transactions.

Meanwhile, blockchain security firm Blockaid highlighted risks associated with AI trading agents, noting that malicious instructions can be embedded in token details, potentially leading to unintended transactions. BlackRock’s Digital Assets Research group also explored the intersection of AI and digital assets, emphasizing the need for native payment methods suited to AI systems capable of multistep tasks with limited human oversight.