On 4 November 2026 the Cyberspace Administration of China, together with other parties at the Wuzhen Summit of the World Internet Conference, issued a "Smart Internet Innovation and Development Cooperation Initiative". It defines the smart internet as the convergence of AI and the internet, and lists eight areas for cooperation. The technical ones: research into intelligent-native network architectures, new transport protocols and compute-network integration, so the network moves from connecting people and information to connecting agents and intelligent services; joint deployment of IPv6, 5G-A/6G, satellite internet and intelligent compute; and international standards for agent interoperability covering interfaces, interconnection protocols and the semantic and process layers, with a trust framework spanning agent discovery, negotiation and execution, described as verifiable and traceable. The rest cover applications (AI phones, humanoid robots, embodied intelligence, model-as-a-service, agent-as-a-service), cross-border data flows under security and compliance, security work against data poisoning and out-of-control agent behaviour with safety and ethics embedded across the development, deployment and operations lifecycle, digital-literacy programmes, and governance dialogue at the UN and other multilateral platforms. The document is an initiative, not a binding rule: it names areas for coordination without setting technical requirements, conformance tests or timelines.
Our takeThe notable move is where the document puts the difficulty: not in model capability but in agent interoperation, naming discovery, negotiation and execution plus a record that can be verified afterwards. Those are protocol questions, and specifying them means choosing a shape for interfaces, always a design commitment. The security clause is the concrete one: data poisoning and agent behaviour are named as things to design against, rather than to audit after deployment.