Mysten’s new AI unit is set to have its own engineering, product development, and other functions in order to reduce dependencies, according to a letter from Cheng. The model for this new unit is based on Mysten’s onchain finance business unit, which was established earlier this year. The focus of the AI unit will be on agent-trust products and AI-native infrastructure.
Cheng sees this handover as a way to free up time for technical leadership while still supporting the Sui and Walrus ecosystem. “This makes my scope more manageable,” he stated in the letter.
Building on Sui and Walrus with AI
The goal of the AI initiative is to enable agents to perform tasks autonomously while staying within user-defined boundaries, as outlined by Cheng. These products and services will be closely tied to Sui and Walrus.
Walrus, Mysten’s decentralized storage protocol, was launched in March 2025, as previously reported by The Defiant.
One of the projects highlighted by Cheng is the Verifiable Agent Arbiter, which was announced by Mysten on October 6. This evidence layer is being developed in collaboration with Google Cloud and is aimed at allowing businesses to demonstrate that their AI agents have operated within their designated authority.
In the proposed system, detailed agent records will be stored in customer-controlled Google Cloud Storage, while cryptographic proofs will be stored on Walrus and managed on Sui. Mysten plans to roll out initial enterprise deployments before making the product more widely available.
