Consider an assistant that books a table, pays a deposit and then cancels on your behalf: the restaurant may have no clear record of who authorised any of it.
That everyday confusion is the problem a new platform unveiled at a Chinese internet conference is meant to address: giving software agents an identity that other systems can recognise and verify.
What Was Unveiled in Wuzhen
On 4 November 2026, a forum on "mutual trust and interoperability" was held at the World Internet Conference in Wuzhen (乌镇), Zhejiang. The session produced several results, the central one being an agent identity registration and resolution platform (智能体身份注册与解析平台).
The platform is designed to solve three technical problems: assigning identity codes to agents, resolving those identities across different domains, and enabling trustworthy interoperability between them. On top of that registry sits an operating platform where agents can be registered and their capabilities published. Ant Group (蚂蚁集团) was named among the first organisations involved in building that operating layer.
Two Ant services were among the first to connect: the government-affairs AI assistant Xiaozheng (晓政) inside Alipay, and Zhima Credit (芝麻信用). Both completed verified connections with agents on other platforms, which the company describes as groundwork for future cross-platform collaboration.
The Three Questions Every Agent Must Answer
Industry participants at the forum framed the problem in plain terms. As agents move from chatting to acting, three questions keep recurring:
- Can a user see what an agent has been permitted to do, and take that permission back in time?
- Can a service provider tell which agent is calling, and on whose behalf it acts?
- Can an agent on one platform find and use a service offered on another?
The first is about consent, the second about identity, and the third about discovery. Each has a technical answer, but only together do they make an agent something a person would delegate real tasks to. A registry that assigns identity codes and resolves them across domains addresses the middle question; user-side authorisation tools address the first.
Ant's contribution extends beyond the registry. The company says it has built APASS, a unified identity account aimed at letting users see and revoke what they have delegated to an agent. On the connection side, it has worked with more than twenty partners on a protocol called ASL and associated trusted-connection capabilities, intended to make identity, intent and authorisation verifiable when agents cross platform boundaries.
Why Ant Group Is in the Middle of This
Ant is not a neutral standard-setter; it is a company with a direct commercial interest. It runs Alipay, a payments and services app used by hundreds of millions of people, and a credit service. If agents begin to transact and act on behalf of users, the company that owns the payment rail and the identity layer has a strong reason to define how that works.
That does not make its participation worthless. The registry is a multi-party effort, and Ant is one participant among several, not the owner of the standard. A company with real consumer traffic can supply something a standards document cannot: a place to test whether a trust mechanism survives contact with actual users. Connecting Xiaozheng and Zhima Credit first is a small but concrete trial of that idea.
Ant's Chen Liang has been quoted as arguing that the agent era requires a trust infrastructure in which identity, authorisation and responsibility travel with every connection — a problem of public infrastructure for the intelligent internet, not merely of product features. He has also argued that the ceiling for agents is set not by whether they can complete a task, but by whether they can be entrusted with one.
What This Does Not Solve
A registry is a starting point, not a guarantee. Naming an agent does not tell you whether it is competent, whether it will follow instructions faithfully, or whether it will act in your interest when instructions are ambiguous. It answers "who is calling," not "can this caller be relied upon."
There is also a governance question the published materials leave open. Who is accountable when a registered agent makes a costly mistake — the user who delegated, the platform that hosted it, or the developer who wrote it? The phrase "responsibility enters every connection" is an ambition, not yet an allocation of liability.
Finally, standards that are built by a handful of large platforms can become barriers rather than bridges if access is priced or restricted. None of the sources reviewed describes an open, permissionless path for small developers, and at the time of writing no published metrics showed how often the cross-platform connections had actually been used.
Key takeaways
- An agent identity registration and resolution platform (智能体身份注册与解析平台) was unveiled at the Wuzhen World Internet Conference forum on 4 November 2026.
- It targets identity coding, cross-domain resolution and trustworthy interoperability, with an operating layer for agents to register and publish capabilities.
- Ant Group joined the first operators; its Alipay government assistant Xiaozheng (晓政) and Zhima Credit (芝麻信用) were among the first services connected.
- Ant says it has built the APASS identity account and, with over twenty partners, the ASL trusted-connection protocol.
- The platform answers "who is calling," not whether an agent is competent or who bears responsibility when it fails.
Honest limitations
This account draws on Ant Group's own press release and on reports from state and party media covering the forum. The event details and the list of connected services are consistently reported across those sources, but they are descriptions of a newly launched initiative, not evidence of widespread adoption. No usage statistics were published in the material reviewed.
The arguments attributed to Chen Liang are his stated views, reported by multiple outlets; they are not independent findings. No currency figures appear in this story, so no RMB-to-US$/HK$ conversion applies.
I have deliberately not treated the launch as a finished standard. The distinction between a registry being unveiled and a trust system being in everyday use is large, and the sources do not close it.
What readers can do now
- Watch what permissions you grant an agent. A trustworthy identity matters little if you have handed over broad authority without noticing.
- Look for revocation, not just identity. The useful test of an agent system is whether you can see and undo what it was allowed to do.
- Treat cross-platform claims as early. A handful of pilot connections is a proof of concept, not proof that the agent internet already works at scale.
