Huawei, Alibaba and Tencent sign a pact on AI agent trust

On 4 November, the China Academy of Information and Communications Technology and 20 groups signed a pact on AI agents. Firms that make devices and cloud services, such as Huawei, signed it at the World Internet Conference in Wuzhen, China. The pact sets 8 measures for agent identity, links between platforms, security and governance, and it is not mandatory.

What the pact says

Agents get identity credentials in tiers that other platforms can check. The signers will make technology for agents to find each other. Agents will use the same protocols, which are rules for how they send data.

For security, the pact wants risk checks, monitors, behavior control and emergency stops. It also wants trusted sources of models, plug-ins (add-on software) and skills. It opposes the use of agents for black-market and gray-market work.

The proof

The source gives no test results, no dates and no standards. The summary does not show the measure on the rights of agent users.

What is not known

The pact is not mandatory. The signers select how far to apply it, and the standards that come after it must give more instructions. It is not known if more firms will sign it.

Insight: Three groups try to set the rules for AI agent trust

In China, large firms signed a pact on agent identity. In October, 3 US agencies released guidance on agent incidents, and 2 US groups announced a joint certification for agents. This reads as a race to set agent rules, since the first to set standards can get an edge. If you sell agents in some regions, different proof of trust is possible in each.