Two AI agents from different vendors meet inside a production system and cannot introduce themselves, prove who they are, or agree on a common language. Multiply that by millions of agents and you get what Chinese standard-setters now call the "agent island" (智能体孤岛) problem—a fragmentation threat to the entire agent economy.
On 26 June 2026, the State Administration for Market Regulation (国家市场监督管理总局, SAMR) approved and released the Artificial Intelligence—Agent Interconnection (《人工智能 智能体互联》) series of seven national standards. They are China's first national-level framework for agent interconnection.
Key takeaways
- The seven standards were formally released (正式通过/批准发布) on 26 June 2026 by SAMR, confirmed by the Standardization Administration (国家标准化管理委员会, SAC).
- They were issued as national standardization guidance technical documents (国家标准化指导性技术文件, GB/Z)—non-mandatory, agile instruments suited to an early industry phase.
- The set spans general architecture, identity code, identity management, agent description, agent discovery, agent interaction and tool calling.
- The explicit aim is to dissolve "agent islands" via unified identity, full-lifecycle traceability and shared interaction rules.
- They are one slice of the "AI Hundred National Standards Build-out" special action (人工智能百项国家标准建设专项行动计划), which plans 146 AI national-standard projects.
- Stage caveat: released as guidance documents, not yet mandatory national standards; the identity-code standard may later be upgraded to mandatory.
What the standards actually do
An autonomous agent—software that perceives, decides and acts toward a goal—is only useful in proportion to the other agents it can talk to. Today, each vendor ships its own interfaces, protocols and identity schemes. The result is duplicated integration work and, worse, no trustworthy way to know whether the agent on the other end is authentic.
The new series attacks the problem at the protocol layer. Rather than mandating a single agent framework, it standardizes the handshake: how an agent is described, found, authenticated, and how it calls tools and talks to peers. Zhu Meina (朱美娜), deputy director of the Standards Technology Management Department at SAMR (国家市场监督管理总局标准技术管理司), described the core problems the standards address as threefold: ecosystem barriers from incompatible interfaces, a trust crisis from missing identity and traceability, and inflated innovation cost from repeated, non-reusable integration.
The closed-loop design
The seven parts form a closed loop: 身份标识 (identity) → 能力描述 (capability description) → 供需发现 (discovery) → 协同交互 (interaction) → 工具调用 (tool calling).
- General architecture sets the overall frame and terminology.
- Identity code (身份码) and identity management (身份管理) give each agent a registrable, traceable digital identity—the "digital ID" officials invoke.
- Agent description (智能体描述) standardizes how an agent advertises what it can do.
- Agent discovery (智能体发现) lets agents and systems find each other by capability.
- Agent interaction (智能体交互) and tool calling (工具调用) define how they cooperate and invoke external functions.
Together they let a developer build once and interoperate across vendors, instead of writing a bespoke connector per platform.
Why identity is the crux
The hardest part of agent safety is not capability; it is accountability. An agent that can act in the world but cannot be identified is a liability. By putting identity code and identity management first, the standards make traceability a precondition of participation rather than an afterthought.
This matters as agents move from chat windows into physical and financial settings. A uniform identity lets a system answer "which agent did this, on whose authority, with what credential?"—the questions any auditor or regulator will eventually ask. China is, in effect, building the interoperability and attribution layer before the agent market fully arrives.
Where this sits in China's AI standards push
The seven are not isolated. SAMR, together with the Ministry of Industry and Information Technology (工业和信息化部), launched the "AI Hundred National Standards Build-out" special action (人工智能百项国家标准建设专项行动计划) in late 2025, planning 146 AI national-standard projects across compute, large models, embodied intelligence, agents and terminals. By a State Council briefing, 15 AI national standards had already been released under the push, including server-performance-test and large-model series.
Crucially, the agent series is issued as GB/Z guidance technical documents (国家标准化指导性技术文件)—a deliberate "use-first, iterate-fast" (急用先行) choice so multiple technical routes can co-exist while consensus forms. Officials signaled the identity-code standard may later be elevated to a mandatory national standard, and that agent-audit and agent-transaction standards are next.
Why this matters beyond China
Agent interoperability is a global coordination problem, not a Chinese one. The European and U.S. ecosystems are converging on their own protocols; China's state-backed, pre-emptive standardization is a different modality—top-down architecture before market lock-in.
For international vendors, the practical question is convergence: will Chinese agent identity and discovery schemes inter-operate with open protocols elsewhere, or become a separate stack? The standards' emphasis on description and discovery (rather than a single mandated runtime) leaves the door open to bridging. Either way, the "digital ID for agents" idea is likely to appear in every jurisdiction's agent-governance debate from here.
Honest limitations
- The seven are guidance technical documents (GB/Z), not mandatory national standards; compliance is currently voluntary, so real interoperability depends on industry uptake we cannot yet measure.
- "Released" does not mean "widely implemented"; we found no public data on how many vendors have adopted the specs in production.
- The standards define protocols, not a conformance regime; without testing tools and certification, interoperability claims remain theoretical.
- Only the identity-code piece is slated for possible mandatory status; the rest may stay guidance indefinitely, limiting enforcement leverage.
What readers can do now
- If you build or buy agents, track the GB/Z texts and any conformance tooling from the China Electronics Standardization Institute (中国电子技术标准化研究院).
- Watch for the agent-audit and agent-transaction standards officials said are next—those will reveal the enforcement direction.
- Compare China's identity-first approach with open agent-protocol efforts in your own market to spot convergence or fragmentation risk early.