Article 1 is an ID card

The convention runs to eight articles, and the order matters. It does not start with safety pledges or industry promotion. It starts with identity. Article 1 commits the signatories to build a trust system for AI agents: hierarchical and verifiable identity identifiers, mechanisms for checking and managing those identities, controls over what each agent is permitted to do, and a way of embedding rules into the agent itself.

Article 2 then names the plumbing: shared infrastructure for agent registration and resolution, capability description and discovery, intent conversion, network routing, and protocol compatibility. Read together, the first two articles describe a phone-book layer for agents. An agent gets an identifier. The identifier can be looked up across domains. The lookup says who operates the agent and what it is allowed to do.

The rest of the text is broader and softer. Article 4 calls for risk assessment, operational monitoring and an abnormal-behavior circuit breaker. Article 5 commits signatories to compliance testing and to resisting the use of agents for algorithmic exploitation and black- or gray-market activity. Article 6 lists user protections: the right to know, the right to choose, final control over important matters, and protection against abuse of personal information and cognitive manipulation.

The enforcement is the name

None of this is a law. The document is a self-discipline convention, and its own preamble says the signatories practice it voluntarily and accept supervision from industry organizations and society. There are no penalties in the text, no deadlines, and no statement of what happens to a company that signs and then does nothing.

That softness should not be read as meaning nothing will happen. CAICT is a research institute under the Ministry of Industry and Information Technology, and the convention states its purpose directly: it implements the industry self-discipline task set out in the government's Implementation Opinions on the standardized application of AI agents. A voluntary commitment of this kind could precede formal requirements; nothing in the documents published so far says that it will.

The convention sits alongside a broader government-backed standardization effort. China published the GB/Z 185 series on agent interconnection in May, covering identity codes, identity management and other interoperability requirements. The Wuzhen convention adds a voluntary industry commitment, while the registration platform provides an early attempt to connect those principles across operating systems. None of these developments, by itself, establishes a compulsory nationwide registration requirement.

A platform that is still a pilot

The registry itself is not finished. According to tech360's reporting from the forum, national-level bodies are building an identity registration and resolution platform jointly, handling identity coding and cross-domain resolution, with operating platforms layered on top to host agent access and publish what each agent can do. The design, by the same reporting, is still being finalized. The platform was unveiled at the summit on November 4, as reported.

What exists is a pilot. Several operating platforms and service providers have been connected, which tests the one property a registry actually needs: resolving an identity issued on someone else's system. According to tech360's November 6 reporting, Ant Group is among the early participants. Its Alipay government-services agent and Sesame Credit have completed trusted links with agents running elsewhere. Chen Liang, an Ant Group vice president and technology chief for its Alipay and security group, framed the point as liability: identity, authorization and responsibility built into every connection. Ant's own APASS unified identity account and its ASL protocol, developed with more than 20 partners, are its contribution to that layer.

The regulator's framing was quoted in the same report. Wang Jingtao, a deputy director of the Cyberspace Administration of China, called for basic rules on agent identity, trusted interconnection, authorization management and compliant operation, and described the shift plainly: from governing content and models to governing verifiable actors, controllable behavior and traceable responsibility.

A separate initiative went cross-border

A separate initiative involving CAICT appeared one day later. Zetrix AI and CAICT announced a Know Your Agent service at Wuzhen and Dubai on November 5, 2026, with a planned cross-border pilot across the Middle East, Africa, China and ASEAN, and findings to be submitted to the ITU and IEEE. The service verifies who an agent represents, what it is authorized to do and whether its actions stay within approved limits.

The sequence still matters. Western agent identity so far is a vendor market: RSA sells agent identity for the enterprise grid, IBM sells an identity vault, Cloudflare sells sites the controls to recognize agents at the door. China is building the same layer as national infrastructure first, with its largest platforms signed up at the start. A company selling agents into China should expect a registration step to arrive before the commercial negotiation does. What nobody can yet say, because the documents do not say it, is when registration becomes compulsory, or what an unregistered agent loses.

China is pursuing agent identity through several connected but distinct efforts: national guidance documents, a voluntary industry convention, and a pilot registration platform. The international Know Your Agent service is a separate initiative, and its cross-border pilot and proposed submission to the ITU and IEEE are future plans, not completed standardization.

Sources

  1. [1] Self-Discipline Convention for AI Agent Development, full text sourced to the China Academy of Information and Communications Technology, released November 4, 2026 at the Wuzhen Summit (eight articles; voluntary wording; the link to the Implementation Opinions on Standardized Application and Innovative Development of AI Agents; CAICT, Tencent, Ant Group, Alibaba and Douyin among the drafting parties, under the organization of bodies including the China Internet Society; read in full)Read source
  2. [2] National Standard Announcement No. 22 of 2026, State Administration for Market Regulation (Standardization Administration of China), May 22, 2026 (approving publication of the GB/Z 185 series, Artificial intelligence — Agent interconnection, parts 1 to 7: overall architecture; identity code; identity management; agent description; agent discovery; agent interaction; agent tool invocation; all seven are national standardization guiding technical documents)Read source
  3. [3] tech360, "China Is Building an Identity Registry for AI Agents", November 6, 2026 (identity registration and resolution platform unveiled at the summit; national-level bodies building it jointly for identity coding, cross-domain resolution and trusted interoperability; design still being finalized; pilot connecting several operating platforms; Ant Group's Alipay government-services agent and Sesame Credit trusted links; Chen Liang's APASS and ASL remarks; Wang Jingtao's framing; image credited to Pandaily)Read source
  4. [4] ChinaTechNews, "Chinese Tech Giants and State Academy Issue National Self-Discipline Pact to Govern AI Agents", November 4, 2026 (20 founding organizations; CAICT as a research institute under the Ministry of Industry and Information Technology; the eight measures)Read source
  5. [5] KLSE Screener, "Zetrix, China's CAICT launch AI agent ID service", November 5, 2026 (Know Your Agent service announced at Wuzhen and Dubai; planned cross-border pilot across the Middle East, Africa, China and ASEAN; findings to be submitted to the ITU and IEEE)Read source