The State Administration for Market Regulation has unveiled a series of national standards focused on the interoperability of artificial intelligence agents, marking a significant step towards a more connected and standardized AI ecosystem. This move is a key part of the systematic and progressive development of China's AI standardization framework.
To implement the strategic decisions of the central leadership for promoting high-quality development in the AI industry, the market regulator, in collaboration with the Ministry of Industry and Information Technology and other departments, is advancing the "Action Plan for the Construction of One Hundred National Standards in Artificial Intelligence." This initiative coordinates 146 national AI standard projects covering key areas such as computing power, large models, embodied intelligence, AI agents, and terminals.
In the foundational support sector, the standard "Artificial Intelligence—Performance Test Methods for Server Systems" has been released. This standard aims to establish performance testing benchmarks for AI computing products, guiding the domestic AI server ecosystem towards scale and collaborative development.
For core technologies, the "Artificial Intelligence—Large Model" series of standards has been issued. These standards unify the entire process of large model development, evaluation, and deployment, solidifying the innovation foundation for core AI technologies.
In the product application domain, the "Artificial Intelligence—Intelligence Grading for Terminals" standard has been introduced. It clarifies the criteria for grading smart terminals, effectively identifying "pseudo-intelligent" products and regulating market order. Simultaneously, it supports the implementation of related industrial support policies and drives the overall intelligent upgrade of the terminal industry chain.
This series of key standards propels AI technology from isolated breakthroughs towards systematic iteration, effectively reducing industry R&D costs and barriers to large-scale implementation, thereby comprehensively supporting the industry's standardized and orderly development.
During a press conference, officials highlighted the rapid evolution of AI, particularly with the advent of large models, which is accelerating its transition from perception and understanding to generation, decision-making, and autonomous execution. AI agents, as intelligent systems with capabilities for autonomous perception, memory, decision-making, interaction, and execution, represent a crucial application form of next-generation AI and a key vehicle for empowering various industries.
While China's AI agent industry is developing swiftly with expanding application scenarios, challenges remain. Issues such as non-uniform communication interfaces, difficulties in interconnection, lack of identity management systems, and unregulated collaborative interaction rules have led to prominent "information silos" among AI agents. Standardization is urgently needed to unify rules, establish order, break down barriers, and mitigate risks.
To address these challenges, seven national standards under the "Artificial Intelligence—Agent Interoperability" series have been approved and released. These standards comprehensively cover core aspects including overall architecture, identity codes, identity management, agent description, agent discovery, agent interaction, and agent tool invocation. This establishes a closed-loop standard system covering "identity identification—capability description—supply-demand discovery—collaborative interaction—tool invocation," effectively filling a gap in this field.
Adopting these unified standards allows companies to reuse standard components, reduce custom development, and shorten product time-to-market. It also establishes unified identity authentication and full-process traceability mechanisms, laying an institutional foundation for secure, cross-domain, and trustworthy interaction.
These standards were initially released as National Standardization Guiding Technical Documents, an agile arrangement during the industry cultivation phase. This approach allows for maximum compatibility with multiple technical routes and rapid consensus-building across the industry, while reserving ample room for technological trial and error. Plans are in place to eventually upgrade the identity code standards to mandatory national standards and accelerate the development of standards in areas like agent auditing and transaction.
Efforts to promote standard implementation involve pooling resources from industry, academia, research, and application. Leading enterprises, research institutes, and application units are encouraged to participate throughout the standard implementation and feedback process. Pilot verification and demonstration projects will be conducted in innovation hubs like Beijing's Haidian District to cultivate standardized, replicable AI agent products and high-value application scenarios.
Officials emphasized that the core function of this interoperability standard series is to "set a baseline, raise the bar, and promote innovation," precisely addressing industry pain points, regulating order, and safeguarding the high-quality development of the AI agent industry. It is an integral part of a broader, systematic push in AI standardization.
Looking ahead, the focus will be on deepening the integration of AI standards with industry, strengthening standard layout, and promoting effective implementation to empower new industrialization and foster new quality productive forces. This includes a demand-driven approach to establish foundational standards, especially in privacy protection, risk control, and security ethics. It also involves fostering ecosystem collaboration to accelerate the application of "AI+" standards in high-value scenarios across industries like manufacturing, agriculture, transportation, and healthcare.
Technological innovation will be prioritized, with a focus on developing standards for AI agents, embodied intelligence, and software-hardware synergy. Activities to promote the adoption of key AI standards will be conducted, selecting benchmark products and solutions where standards empower the industry. Furthermore, China will actively engage in international standardization, contributing its technological achievements and mature practices to global AI governance, offering a "Chinese approach."
The seven parts of the interoperability standard series are designed with the logic of enabling "orderly and trustworthy collaboration among AI agents," forming a complete and logically closed technical foundation. The standards address everything from top-level architecture and establishing digital identities for agents to describing their capabilities, enabling discovery, and defining interaction and tool-calling rules.
To ensure practical application, a tripartite model of "joint standard research, open-source development, and adaptation evaluation" is being promoted. This includes launching a "Pioneer Plan" for standard application, advancing the development and iteration of open-source code for the Agent Interoperability Protocol (AIP), and developing specialized evaluation toolkits based on established national AI benchmarking systems.
Overall, these standards transform isolated "islands" into interconnected "networks," turn proprietary systems into open ecosystems, and bring order to potential chaos. By setting basic principles and capability thresholds, they provide foundational support for the safe, trustworthy, and controllable development of AI, achieving a dynamic balance between industrial innovation and security governance.
Regions with strong AI industrial bases, such as Beijing's Haidian District, are positioned to take the lead in piloting these national standards. Leveraging their industrial and scenario advantages, they will accelerate ecosystem cultivation, deepen scenario empowerment across key sectors, and strengthen policy linkages to reduce the barriers to standard application and output replicable models for standardized, safe, and collaborative AI agent development.
The work on AI standardization is entering a new phase of acceleration and quality improvement. Beyond AI agents, future national standards in development or planned cover frontier technologies like embodied intelligence and world models, foundational elements such as computing infrastructure and high-quality datasets, industry-specific large model applications, and AI safety standards for critical sectors like finance and healthcare.
Long-term plans involve a coordinated layout across the entire chain of AI standards—from foundational and core technologies to industry applications and safety governance. This includes clarifying standard boundaries in overlapping fields, deepening the linkage between scientific innovation and standardization, and optimizing the entire standard management process to enhance efficiency and ensure standards remain timely and applicable amidst rapid technological change.
Comments