Sui Markets
Loading prices…
SUIWIRE

Mysten Labs and Google Cloud Plan a Verifiable Record for AI Agents

The Verifiable Agent Arbiter would log what an AI agent was allowed to do and what it actually did, with proofs anchored on Walrus and Sui.

SuiWireNewsroom1 min read
Mysten Labs and Google logos
Logos: Mysten Labs, Google

Mysten Labs and Google Cloud announced plans on October 6 for a platform that checks whether AI agents stay within the limits their users set. The system is called the Verifiable Agent Arbiter.

How it works

As agents start to handle banking, payments and service requests, companies need a way to review what an agent was authorized to do and what it actually did. According to TokenPost and The Herald Business, the Arbiter records an agent's authorizations, actions and results so a company can audit how it handled a transaction or request.

The design splits the data. Instructions, model outputs and tool-call records stay in Google Cloud storage. Cryptographic proofs of those records are stored on Walrus, the decentralized storage network in the Sui ecosystem, and verified through Sui. The companies also plan to connect the Arbiter to agent payment services, so service calls, payments and authorization records are linked.

Rollout

Enterprise deployment comes first, followed by a wider release. Neither company has published pricing, launch customers or a firm date. TokenPost noted that the announcements did not disclose the expected scale of enterprise payments or fees.

Why it matters for Sui

The Arbiter fits Sui's push to become the settlement and verification layer for what it calls agentic commerce. That push also includes Sui Agent Payments, the Alibaba Cloud collaboration announced the next day, and the tunnels throughput demo at Basecamp.

The token economics are indirect. Storage on Walrus is paid in WAL, and onchain operations on Sui are paid in SUI. How much activity an enterprise audit trail would actually generate is still unknown.

Mysten Labs also framed the work as a security effort. In comments to The Herald Business, co-founder Adeniyi Abiodun pointed to the speed at which AI agents could attempt attacks as a new class of risk for financial systems.

Sources