Cobo proposes the "Contracted Permissions + MPC" framework: building a controllable trust layer for the Agent economy

BlockBeats News, April 21 — At an offline event themed “Decoding Web 4.0: When AI Agents Take Over On-Chain Permissions,” Brad Bao, Cobo AI’s Head of Growth, said that Cobo is reconstructing an Agent’s fund control logic through a structured permission system. In its products, the relationship between users and Agents is no longer a simple wallet authorization relationship; instead, permission boundaries are defined by an executable “contract (Policy/Agreement).”

This contract includes at least four core elements: Intent, Path, Rules, and Completion and Termination conditions, thereby upgrading permission control from the original “all or nothing” model into a fine-grained governance mechanism that is negotiable, auditable, and revocable.

At the execution level, Cobo combines MPC (Multi-Party Computation) technology to ensure that even if the Agent exhibits abnormal behavior, the system is attacked, or risks such as prompt injection occur, the funds are always controlled by the user, and no single party can bypass the constraints to transfer assets independently. Brad Bao emphasized that under this architecture, the Agent can obtain sufficient operational freedom to complete complex tasks, but its actions are always limited within verifiable boundaries.

He said that, in essence, this system builds a layer of “trust and insurance infrastructure” within the Agent economy, and by pairing “contractualized permissions + secure computation,” it enables controllable delegation of Agent behavior. Ultimately, while improving automation efficiency, it also safeguards users’ asset security and ensures result certainty.