AI 监管 · 政策

首套智能体互联国标发布
覆盖5大维度7项文件

First National Agent Interconnection Standards
7 Documents Covering 5 Dimensions

国家市场监管总局发布GB/Z185-2026系列7项国标,覆盖身份标识、能力描述、供需发现、协同交互、工具调用全链路,AIP开源项目已吸引华为、360等百余家企业参与。

China's SAMR publishes GB/Z 185-2026 series of 7 national standards covering identity, capability description, discovery, collaboration, and tool invocation; open-source AIP project already has 100+ companies including Huawei and 360.

No.012 2026.07.10 约 5 分钟阅读 ~5 min read

7项文件,5大维度,100+企业参与。中国在AI Agent协议标准化上,先于欧美落下了第一颗棋子。

国家市场监督管理总局、国家标准化管理委员会正式批准发布GB/Z 185-2026《人工智能 智能体互联》系列7项国家标准化指导性技术文件。这是国内首套覆盖智能体全生命周期的国家级标准体系,标志着中国在Agent互联互通这一关键赛道上,从市场应用领先走向标准制定领先。

五大维度覆盖Agent全链路

GB/Z 185系列标准的架构非常完整,覆盖了一个AI Agent从"出生"到"协作"的全链路。

标准围绕五大核心维度展开:身份标识、能力描述、供需发现、协同交互、工具调用。身份标识解决"我是谁"的问题,给每个Agent一个唯一可识别的身份证;能力描述解决"我能干什么"的问题,定义Agent能力的结构化描述方式;供需发现解决"谁能帮我干活"的问题,建立Agent服务的注册和发现机制;协同交互解决"我们怎么配合"的问题,定义多Agent之间的通信协议和协作模式;工具调用解决"我怎么动手"的问题,统一Agent调用外部工具和资源的接口。

这五个维度基本上把Agent互联的所有问题都覆盖了。对比国际上正在发展的协议生态——MCP解决工具调用、A2A(Agent-to-Agent)解决协同交互、ACP解决IDE内Agent接入——中国的国标是一个更顶层、更全面的框架,试图把各种协议纳入统一的标准体系下。

标准的工程化实现是AIP(Agent Interconnection Protocol)智能体互联开源项目,由北京邮电大学和中国电子技术标准化研究院联合研发。AIP v2.1.0刚刚发布,已经吸引了华为、360等百余家企业参与。一个由学术界牵头、产业界广泛参与的开源标准生态,已经初具雏形。

标准制定权就是产业话语权

为什么这套国标值得关注?因为在AI产业进入Agent时代的关口,标准制定权就是产业话语权

回顾互联网的发展史,TCP/IP协议定义了网络层的通信规则,HTTP协议定义了Web层的交互标准——谁制定了协议,谁就掌握了生态的主导权。在AI Agent时代,全球还没有形成统一的互联标准:美国方面,MCP由OpenAI主导、A2A由Google主导、ACP由JetBrains等社区推动,各自为政;欧盟还在聚焦AI法案的监管落地,在技术标准上动作缓慢。

中国在这个时间窗口推出统一的国家级标准,至少释放了三个信号。第一,中国的Agent产业应用已经走到了需要标准的阶段——当市场上有成千上万个Agent在跑,互联和互通就成了刚需,标准是产业成熟的标志。第二,中国试图在Agent互联标准上抢得先机——国标先行,再推动国际标准采纳,这和5G标准制定的策略一脉相承。第三,监管和发展并行——这套标准不仅是技术规范,也为后续的Agent治理、安全审计、责任认定提供了技术基础。

当然,指导性技术文件(GB/Z)和强制性国家标准(GB)的法律约束力不同——GB/Z是推荐性的,不强制要求执行。但在中国的产业环境中,推荐性国标往往会成为政府采购、国企招标、行业准入的事实标准,其实际影响力不可小觑。

Agent协议的战争,才刚刚开始。MCP、A2A、ACP、AIP——谁能成为Agent时代的"HTTP",谁就能定义下一个十年的AI基础设施格局。中国已经亮出了自己的牌。

明天见。

Seven documents, five dimensions, 100+ participating companies. China has placed its first piece on the AI agent protocol standardization chessboard, ahead of Europe and the US.

China's State Administration for Market Regulation and Standardization Administration formally approved the GB/Z 185-2026 "Artificial Intelligence — Agent Interconnection" series of seven national standardization guidance technical documents. This is China's first national standards framework covering the full agent lifecycle, marking China's transition from market application leadership to standard-setting leadership in this critical arena.

Five Dimensions Covering the Full Agent Stack

The GB/Z 185 series has a remarkably comprehensive architecture, covering the full chain from an AI agent's "birth" to its "collaboration."

The standards revolve around five core dimensions: identity, capability description, supply-demand discovery, collaborative interaction, and tool invocation. Identity solves the "who am I" question by giving each agent a unique identifier. Capability description solves "what can I do" by defining structured representations of agent abilities. Discovery solves "who can do this for me" through agent service registration and lookup mechanisms. Collaboration solves "how do we work together" by defining inter-agent communication protocols and coordination patterns. Tool invocation solves "how do I take action" by unifying interfaces for agents calling external tools and resources.

These five dimensions cover essentially every aspect of agent interconnection. Compared to the evolving international protocol ecosystem — MCP handling tool invocation, A2A (Agent-to-Agent) handling collaboration, ACP handling in-IDE agent integration — China's national standard is a higher-level, more comprehensive framework that seeks to bring various protocols under a unified standards umbrella.

The engineering implementation is AIP (Agent Interconnection Protocol), an open-source project led by Beijing University of Posts and Telecommunications and the China Electronics Standardization Institute. AIP v2.1.0 just shipped, with 100+ companies including Huawei and 360 already participating. An open standards ecosystem led by academia with broad industry participation is taking shape.

Standard-Setting Power Is Industry Clout

Why does this national standard matter? Because as the AI industry enters the agent era, whoever sets the standards holds the industry's microphone.

Looking at internet history, TCP/IP defined network-layer communication rules and HTTP defined web-layer interaction standards — whoever made the protocol controlled the ecosystem. In the AI agent era, no unified interconnection standard exists globally: in the US, MCP is led by OpenAI, A2A by Google, ACP by JetBrains and community efforts — each doing its own thing. The EU remains focused on implementing the AI Act, moving slowly on technical standards.

China launching a unified national standard in this window sends at least three signals. First, China's agent application market has reached the stage where standards are necessary — when thousands of agents are running in the market, interconnection and interoperability become essential, and standards are a sign of industry maturity. Second, China is trying to get a head start on agent interconnection standards — national standards first, then pushing for international adoption, following the same playbook as 5G standard-setting. Third, regulation and development move in parallel — these standards aren't just technical specifications; they provide a technical foundation for future agent governance, security auditing, and liability determination.

Of course, guidance technical documents (GB/Z) don't carry the same legal force as mandatory national standards (GB) — GB/Z is recommended, not required. But in China's industrial environment, recommended national standards often become de facto standards for government procurement, SOE bidding, and industry entry, giving them practical clout.

The agent protocol wars have only just begun. MCP, A2A, ACP, AIP — whichever becomes the "HTTP" of the agent era will define the next decade's AI infrastructure landscape. China has played its card.

See you tomorrow.

GB/Z185-2026 · 智能体互联 · 国家标准 · 7项文件 · 5大维度 · AIP开源项目 · 北京邮电大学 · 华为 · 标准制定权 · Agent协议战争
GB/Z 185-2026 · agent interconnection · national standard · 7 documents · 5 dimensions · AIP open source · BUPT · Huawei · standard-setting · agent protocol wars
Sources · 信源 Sources

本文基于 Dawn Vision 认知引擎处理的 11 个源信号生成,素材来源:人民网、51CTO、今日头条。

This article was generated from 11 source signals processed by the Dawn Vision cognitive engine. Sources: People's Daily, 51CTO, Toutiao.