China's first policy framework for AI agents.
On May 8, 2026, China’s Cyberspace Administration, National Development and Reform Commission, and Ministry of Industry and Information Technology jointly released the Implementation Opinions on the Standardized Application and Innovative Development of Intelligent Agents.
The document is highly significant because it marks the first time China has begun systematically treating “Agentic AI” as a future digital infrastructure and governance object in its own right, rather than simply another application layer built on top of large language models.
Over the past few years, Chinese AI regulation has focused mainly on generative AI, model filing requirements, content moderation, and data compliance. But this document clearly shows that regulators now recognize something important: intelligent agents are fundamentally different from traditional chatbots. They are not just systems that “generate content.” They are beginning to possess autonomous perception, long-term memory, tool use, cross-platform task execution, and even multi-agent coordination capabilities. In other words, AI is moving from “answering questions” toward actually “doing things” on behalf of users.
At its core, the document is trying to answer a simple but profound question:
If large numbers of intelligent agents truly enter the real world, how does China want them to exist inside society?
Overall, the approach reflects a very typical Chinese industrial-policy mindset: acknowledge the risks, but make absolutely clear that China does not intend to miss the industrial opportunity. The goal is to push “development” and “governance” forward at the same time.
That is why the document immediately sets an important tone. It explicitly states that intelligent agents will deeply integrate into both cyberspace and the physical world, reshaping production, daily life, and social governance. Implicitly, this means Chinese regulators already see Agentic AI not as a niche technology, but as a future foundational layer of the digital economy.
At the same time, the document repeatedly emphasizes ideas like “safe and controllable,” “reliable and trustworthy,” and “orderly governance.” This reflects a broader concern in China today: not simply that AI may become “too smart,” but that once agents gain real autonomous execution capabilities, they could begin exerting systemic effects on society itself.
As a result, the document spends a surprising amount of time discussing decision boundaries, behavioral controls, risk warnings, traceability, human oversight, and multi-agent coordination. Reading it, you can clearly sense that Chinese regulators are already thinking seriously about a future where AI is no longer just a tool, but gradually evolves into something closer to a semi-autonomous digital actor.
One especially important section concerns “decision authority.” The document proposes distinguishing between actions that must always remain under direct human control, actions that may be delegated to AI through authorization, and actions agents may perform autonomously. At the same time, it stresses that users must always retain the right to know and the final right to decide. In some ways, this resembles Europe’s ongoing discussions around “meaningful human control.”
And behind this lies one of the core questions in agent safety:
Can agents make decisions for humans? And if so, to what extent?
Once you translate that into real-world products, the implications become very concrete. Can an agent place orders automatically? Make payments? Delete files? Send emails to customers? Submit government forms? Approve loans? Control industrial equipment? What once looked like ordinary product features may increasingly become regulatory questions.
Another particularly notable aspect is the document’s concern about anthropomorphism and emotional dependency. It explicitly warns against agents using human-like interaction techniques to create addiction, emotional attachment, or manipulative consumer behavior among minors and elderly users. This suggests Chinese regulators are no longer viewing AI simply as an information tool, but increasingly as a social technology capable of reshaping relationships, psychology, and human behavior.
The document also contains another extremely important signal: the idea of an “Intelligent Internet.” It discusses research into intelligent internet architecture, agent registration platforms, digital identities for agents, capability declarations, and multi-agent interoperability protocols. It even explicitly mentions AIP and IPv6.
At a certain point, this stops sounding like regulation for individual AI products and starts sounding more like planning for a future internet where AI systems interact directly with one another. China already appears to be thinking ahead about how agents will communicate, authenticate identity, exchange permissions, make payments, assign responsibility, and establish trusted coordination mechanisms. In some ways, this resembles the early foundations of an “Agent Internet.”
On the industrial side, the scope of the document is extremely broad. It covers almost every major sector imaginable: manufacturing, energy, finance, healthcare, education, transportation, government services, judicial systems, public security, scientific research, and more. The logic is very clear: China does not want intelligent agents to remain at the chatbot stage. It wants them deeply embedded into the real economy and governance systems, becoming the operational layer of the country’s broader “AI Plus” strategy.
This is especially visible in the sections on industrial governance, public administration, and social management. The document discusses policy recommendation systems, AI-assisted approvals, smart judicial systems, public opinion guidance, emotional intervention systems, risk warning mechanisms, and urban governance. The overall flavor feels very much like a distinctly Chinese approach to digital governance.
At the same time, the document strongly emphasizes “indigenous controllability.” It specifically highlights open-source frameworks, compatibility with open-source chips and domestic operating systems, local ecosystem development, and participation in international standards-setting. This is entirely consistent with China’s broader strategy over the past several years in semiconductors, operating systems, and industrial software.
The underlying logic is straightforward:
China does not want the next generation of AI infrastructure to depend entirely on foreign technology stacks.
But the document also explicitly states that China intends to “actively participate in international standards-setting” for intelligent agents. This is important because it signals that China does not want to merely follow standards defined elsewhere — particularly by the United States — but wants to help shape future global protocols and governance frameworks itself.
That matters because global competition around agent standards is already beginning to emerge. The Linux Foundation, for example, has already launched an industry association focused on Agentic AI interoperability and governance standards, including protocols for computation, function calling, identity systems, and authorization mechanisms. The initiative is reportedly open and welcomes Chinese participation. Huawei recently joined as a Gold Member, alongside Lenovo. The association also plans to work with organizations like W3C and IETF in order to gradually evolve today’s relatively experimental and industry-driven standards into globally recognized protocols.
China’s active participation in these international standards discussions is extremely important because agent governance ultimately cannot rely solely on domestic law. Intelligent agents are inherently cross-platform, cross-system, and cross-border. In the long run, the most important issue may not be regulation of individual applications, but rather the interoperability protocols between agents, tools, and platforms.
If China and the United States eventually develop completely incompatible agent ecosystems, the global AI environment could begin fragmenting in ways reminiscent of the splintering of the internet itself. Even if fully unified global standards prove impossible, there will likely still need to be at least some basic consensus around safety protocols, logging and traceability, permission management, and risk controls.
Another interesting aspect of the document is what it does not prioritize. Unlike many American AI safety discussions, it does not place “catastrophic risk” at the absolute center. Of course, it still discusses loss of control, cyberattacks, criminal misuse, and supply-chain security. But the overall emphasis remains on industrial deployment, social governance, application safety, behavioral boundaries, and compliance systems.
More concretely, the document proposes a classification-based governance model depending on application scenarios and potential impact. High-risk sectors will face much stronger regulation, while low-risk scenarios will rely more heavily on platform governance and industry self-regulation. Sensitive sectors and key industries will face filing requirements, testing, product recalls, and oversight by both cyberspace regulators and sector-specific authorities. Meanwhile, lower-risk consumer and office-use scenarios will rely more on self-assessment, reporting systems, platform management, and industry norms.
This will likely have direct implications for product planning. A general office-work agent may primarily need user authorization systems, privacy protections, logging, and enterprise controls. But a financial risk-control agent, medical diagnosis assistant, government approval agent, judicial assistant, or public-security agent will likely require far stricter filing, testing, certification, auditing, recall, and liability frameworks.
In other words, market access costs for intelligent agents will increasingly become tightly linked to scenario risk. The closer an agent gets to finance, healthcare, government services, judicial systems, transportation, energy, or public security, the less companies will be able to compete purely on model capability alone. They will also need to demonstrate safety architecture, compliance processes, testing documentation, third-party evaluations, and traceability.
There is also a clear difference between Chinese and American thinking on agent risk itself. Many U.S. experts worry about autonomous loss-of-control scenarios, especially the possibility that agents could bypass resource limits, autonomously participate in online transactions or prediction markets, or find gray-market ways to acquire additional resources.
Many Chinese experts, by contrast, see these scenarios as relying on unrealistic assumptions of “unlimited resources and unlimited permissions.” In practice, they argue, humans can still impose hard constraints through compute quotas, credit ceilings, access permissions, and system shutdown mechanisms. As a result, many in China believe catastrophic narratives about agentic AI are sometimes overstated.
In that sense, the document’s overall approach to risk governance reflects a broader mindset currently common across China’s policy and industrial circles. China’s approach to Agentic AI today is less “pause first, govern first,” and more “deploy first, govern along the way.” Compared with distant superintelligence scenarios, China’s current priorities are more practical: how to integrate AI into the real economy and society, how to avoid social disorder, how to preserve governance capacity, how to build industrial ecosystems, and how to ensure the entire process remains controllable from beginning to end.
Full translation of the Implementation Opinions on the Standardized Application and Innovative Development of Intelligent Agents.
Agents are intelligent systems capable of autonomous perception, memory, decision-making, interaction, and execution, and represent an important form of artificial intelligence products and services. With the rapid advancement of next-generation AI technologies such as large models, agents are increasingly integrating deeply with cyberspace and the physical world, profoundly transforming modes of production, daily life, and social governance. To implement the State Council’s Opinions on Deepening the “AI+” Initiative and to promote the standardized application, innovation, and development of agents, these Implementation Opinions are hereby formulated.
I. Fundamental Principles
Guided by the objectives of advancing technological innovation, enhancing governance capacity, building industrial ecosystems, and improving public welfare, efforts shall adhere to the principle of safety and controllability, taking the safety, reliability, and trustworthiness of agents as baseline requirements throughout the entire lifecycle of agent technology R&D, deployment, and commercialization, so as to effectively prevent systemic risks. Development shall remain standardized and orderly, adapting to the evolving characteristics of agent technologies, establishing a governance framework that aligns with existing laws and regulations, encourages industry self-governance, and clearly defines bottom lines and red lines, thereby promoting the orderly deployment of agents. Innovation-driven development shall be upheld by strengthening the integration of theoretical, technological, and engineering innovation, systematically advancing breakthroughs in key agent technologies, improving collaborative mechanisms among government, industry, academia, research institutions, and users, fostering an open and shared agent ecosystem, and enhancing industrial innovation vitality. Application-driven development shall focus on practical needs in scientific research, industrial development, consumption promotion, public welfare, and social governance, leveraging the demonstrative role of representative application scenarios and promoting gradual, step-by-step implementation to facilitate technology validation, product iteration, and real-world deployment.
II. Strengthening the Foundations for Development
Efforts shall be made to reinforce technological foundations, improve standards systems, and lower barriers to agent development, adaptation, and application, thereby laying the groundwork for a richer ecosystem of agent products and services.
(1) Improving the Technological Foundation
1. Strengthening Basic Technology R&D
Continue improving the performance of general-purpose foundation models while supporting the development of domain-specific models for specialized industries, forming a diversified model product matrix adaptable to different scenarios and devices. Enhance the supply of high-quality datasets for agent training and operation. Strengthen research on agent task understanding, task planning, tool usage, long-term memory, interoperability, and multi-agent collaboration to improve generalization capabilities.
2. Improving the Agent Toolchain
Conduct research on underlying agent frameworks and accelerate the development of key components for perception, memory, decision-making, interaction, and execution. Improve toolchains for agent development, testing, deployment, and operations. Develop security and governance tools such as adversarial sample detection and abnormal behavior detection to enhance capabilities for discovering, intervening in, blocking, and recovering from non-compliant agent behavior.
(2) Establishing Standards and Protocols
3. Establishing an Agent Standards System
Develop guiding documents for agent standardization work and establish an overall standards framework covering key technologies, major products, data exchange, application scenarios, quality evaluation, security assurance, and trustworthy certification. Accelerate the formulation of foundational standards for interfaces between agents and software tools, application services, and hardware peripherals. Strengthen the promotion and application of key national and industry standards related to agent interoperability protocols (AIP) and other interconnection technologies. Support the formulation of mandatory standards in sectors such as healthcare, transportation, media, and public security. Encourage enterprises to develop products and services in accordance with relevant standards to improve compliance and standardization. Actively participate in the development of international standards.
4. Planning and Developing the Intelligent Internet
Study and establish the architecture of an intelligent internet system. Explore the creation of agent registration platforms providing services such as digital identity management, discovery and search, and capability declarations for agents, while supporting information inquiry and management related to developers, deployment methods, interface protocols, and compliance certifications. Improve multi-agent collaboration capabilities and research foundational technologies including agent identity systems, trusted interconnection, compliant payments, security protection, and conflict resolution. Leverage the advantages of Internet Protocol Version 6 (IPv6) to improve end-to-end communication capabilities for agents. Explore the establishment of monitoring and evaluation indicators for the intelligent internet.
III. Safeguarding Security Baselines
Adhere to a people-centered and AI-for-good approach featuring multi-stakeholder governance and prudent security management, creating an institutional environment that both regulates development and encourages innovation, thereby ensuring the healthy and orderly development of agents.
(1) Clarifying Product Principles
5. Improving Policies, Regulations, and Ethical Norms
Accelerate research into policies, regulations, and ethical standards related to agents. Leverage the advantages of professional institutions in content resources and review mechanisms to ensure agent behavior complies with laws, regulations, and mainstream societal values. Prevent agents from exploiting data advantages or anthropomorphic technologies to spread harmful values or engage in algorithmic exploitation, and guard against risks such as addiction and emotional dependence among minors and the elderly. Ensure coordination with AI ethics review mechanisms and related systems.
6. Clarifying Decision-Making Authority
Under the premise of complying with laws, regulations, social morality, and ethical norms, clearly define the boundaries and required permissions for decisions reserved exclusively for users, decisions requiring user authorization, and autonomous agent decisions. Ensure users retain the right to know and the ultimate decision-making authority over autonomous agent actions, and ensure agents do not exceed the scope of user authorization.
7. Strengthening Behavioral Controls
Develop technologies such as embedded rules and behavioral guardrails to ensure agents operate lawfully and compliantly in public, private, and specialized environments. Explore the use of blockchain and related technologies to establish verifiable and traceable mechanisms for agent behavior in key application scenarios, thereby preventing major risks arising from improper agent conduct.
(2) Preventing Security Risks
8. Enhancing Intrinsic Security Capabilities
Research security technologies related to agent data security, personal information protection, cryptographic safeguards, attack detection, permission management, and behavioral controls. Improve the security assurance capabilities of agent systems and guard against risks such as data poisoning, privacy leakage, algorithm tampering, system vulnerabilities, and loss of operational control. Study agent security testing technologies and explore the establishment of an agent security evaluation system.
9. Strengthening Supply Chain Security
Develop full lifecycle security standards covering agent development, deployment, application, and maintenance. Strengthen security management in areas such as model access, API calls, and the use of extension tools. Explore the establishment of information-sharing and early-warning mechanisms for agent supply chain security, publish timely risk alerts, and improve overall security assurance capabilities.
10. Mitigating Application-Derived Risks
Improve regularized risk identification, early warning, and intervention mechanisms for agents, while strengthening human-machine collaborative review, interception, and blocking capabilities to prevent systemic security risks. Strengthen security management of agent applications to prevent agents from being used for automated attacks, privacy violations, false information generation and dissemination, online fraud, and other illegal or criminal activities.
(3) Improving the Governance System
11. Building a Tiered and Classified Governance Framework
Based on application scenarios and potential impacts, prudently implement tiered governance for agents. For sensitive sectors and key industries, cyberspace authorities together with relevant regulators shall determine permissible application scenarios and implement measures such as filing requirements, testing, and product recalls in accordance with laws, regulations, regulatory requirements, and security standards. For lower-risk fields such as daily entertainment and office work, improve agent evaluation and testing tools and achieve efficient governance through compliance self-assessment, information reporting, platform governance, and industry self-regulation.
12. Improving Compliance Service Systems
Strengthen the supply of professional services related to risk monitoring and warning, testing and evaluation, consulting, and certification for agents, and encourage the industry to develop agent monitoring tools. Carry out third-party testing and evaluation services for agent functionality, performance, quality, and compliance, promote mutual recognition of certification and testing results, and provide users with references for selecting agents. Compile and publish reports on the technological and application maturity of agents to support industrial R&D and deployment.
(4) Strengthening Industry Self-Regulation
13. Guiding Industry Self-Discipline
Encourage industry associations and major enterprises to jointly formulate self-regulatory rules covering agent functionality compliance, algorithm governance, intellectual property protection, and fair competition. Guide agent development platforms, distribution platforms, and service providers to establish fair and reasonable platform rules, user agreements, and privacy policies that clearly define the rights and responsibilities of both supply and demand sides, thereby safeguarding healthy industry development. Strengthen public education on agent application risks and improve user safety awareness.
14. Exploring Credit Evaluation Mechanisms
Guide industry organizations in establishing voluntary credit evaluation mechanisms for market participants in the agent ecosystem. Conduct credit evaluations and lawful disciplinary actions against behaviors such as technology abuse, induced consumption, false advertising, and concealment of defect information. Encourage developers, development platforms, distribution platforms, and service providers to participate in credit evaluation systems and jointly foster a healthy development environment.
IV. Strengthening Application-Driven Development
Promote the steady and prudent deployment of agents in representative application scenarios to drive improvements in technology and products, while exploring replicable and scalable deployment models.
(1) Scientific Research
15. Research Exploration
Develop agents for theoretical reasoning and simulation to explore potential technological pathways. Strengthen agents’ capabilities in information association, integration, and knowledge system construction to improve discovery capabilities in natural sciences, philosophy, and social sciences. Promote integration between agents and scientific instruments and experimental platforms to enable intelligent end-to-end processes covering experimental design, execution, data processing, and result analysis.
16. R&D Assistance
Develop software engineering agents capable of improving the full development lifecycle including requirements analysis, architecture design, code generation, and testing. Promote integration between agents and software such as Computer-Aided Design (CAD) and Computer-Aided Engineering (CAE), enabling functions such as design generation, simulation validation, and parameter optimization.
(2) Industrial Development
17. Intelligent Manufacturing
Develop production management agents capable of dynamically optimizing scheduling, resource allocation, and process coordination, and promote the application of agents in the industrial internet to improve enterprise management efficiency. Enhance capabilities in process parameter optimization, precision inspection, and defect identification, and integrate agents with CNC machines, industrial robots, and automated production lines to improve quality, efficiency, and cost control.
18. Energy and Resources
Develop agents for sensing environmental factors such as atmosphere, water, soil, and noise to improve early warning capabilities for natural disasters and environmental pollution. Use agents to strengthen full lifecycle management of territorial and spatial resources. Utilize agents for efficient exploration of energy and mineral resources such as metals. Develop agents for power dispatching, electricity monitoring, and grid maintenance to improve energy utilization efficiency.
19. Transportation
Develop agents for traffic safety supervision and emergency command and dispatch to improve capabilities in detecting violations, warning of infrastructure risks, supervising key vehicles and vessels, and rapidly responding to accidents. Optimize transportation monitoring and dispatch agents and develop agents for transportation vehicle management to improve the efficiency of road networks, waterways, and airspace.
20. Agricultural Production
Develop agricultural service agents providing agricultural technical guidance, pest diagnosis, and prevention services. Promote the application of agents in planting, breeding, and efficient breeding technologies to accelerate agricultural modernization. Promote integration between agents and smart agricultural machinery, intelligent greenhouses, and agricultural service platforms to improve agricultural productivity.
21. Financial Services
Develop financial risk-control agents to improve risk identification in credit approval, transaction monitoring, and account security. Enhance anomaly detection and compliance auditing capabilities to improve credit default prediction, credit card fraud interception, and anti-money laundering monitoring.
(3) Stimulating Consumption
22. Terminal Applications
Promote the empowerment of internet applications and services through agents to optimize user experiences in online shopping, navigation, bill payment, and office work. Promote coordinated development between agents and devices such as smartphones, computers, automobiles, smart homes, wearables, and consumer robots, enhancing cross-application and cross-device task execution capabilities.
23. Culture and Tourism
Develop agents for literature, music, painting, audiovisual production, and performing arts content creation to promote the dissemination of outstanding culture. Develop tourism service agents for intelligent guidance, multilingual translation, elderly accessibility, and disability assistance to improve tourism services.
24. Commercial Services
Enhance customer service agent capabilities to provide 24/7 consultation, reservations, and after-sales services. Develop embodied agents for guidance, cleaning, warehousing, and distribution to improve operational efficiency in restaurants, retail, hospitality, and logistics. Explore the use of embodied agents to provide low-cost domestic services, elderly care, childcare, and disability assistance.
(4) Public Welfare
25. Education
Explore agents for courseware generation, assignment grading, and learning analytics to improve teacher efficiency. Use agents to develop personalized learning plans and improve intelligent tutoring, Q&A support, and virtual teaching assistant functions. Support online education platforms in developing agents to provide lifelong learning services.
26. Healthcare
Enhance the performance of medical assistance agents in medical imaging analysis, disease diagnosis reasoning, and customized treatment plan generation. Explore agents for pharmaceutical management, surgery scheduling, and medical record management to improve healthcare efficiency. Gradually develop pre-diagnosis consultation and report interpretation agents to improve patient experience.
27. Human Resources
Explore the use of agents in employment promotion, technical skills training and evaluation, and labor relations public services to improve employment support capabilities. Develop agents for social insurance, labor dispute arbitration, and wage arrears governance to safeguard workers’ lawful rights and interests.
28. Information Services
Explore the use of agents in online content governance and encourage information publishing departments and content platforms to develop agents for user analysis, topic planning, content editing, recommendation distribution, intelligent moderation, public opinion guidance, emotional counseling, and real-time translation, thereby enabling efficient integration of multimodal and cross-domain information.
(5) Social Governance
29. Government Services
Explore agents for assisted administrative approval to improve the intelligence of government approval processes. Develop policy consultation agents providing around-the-clock government consultation and procedural guidance services. Explore proactive delivery of policy recommendations, service reminders, and application guidance, accelerating the shift from “people searching for services” to “services finding people.”
30. Judicial Services
Explore full-process judicial assistance agents to improve capabilities in organizing case materials, entering case information, reviewing evidence, and generating legal documents. Develop agents for legal publicity, consultation, and supervision to provide efficient and convenient online judicial services for the public.
31. Public Safety
Explore agents for monitoring and warning, emergency response, rescue dispatch, and coordinated governance to improve capabilities in workplace safety supervision and disaster prevention, mitigation, and relief. Improve abnormal behavior recognition, threat warning, and dynamic prevention and response capabilities to safeguard public security. Promote the deployment of embodied agents in disaster rescue, security inspection, and hazardous materials handling.
32. Urban Governance
Explore the application of agents in urban planning, construction, and governance to support intelligent construction, building management, and safe operation of urban infrastructure, thereby improving the professionalism of urban governance and enhancing urban living environments.
33. Tendering and Bidding
Explore tendering and bidding agents to achieve intelligent end-to-end management of bidding activities and ensure standardized and efficient processes. Improve the intelligence of bidding transactions, services, and supervision to ensure scientific tendering, fair evaluation, and efficient regulatory oversight.
V. Building an Innovation Ecosystem
Facilitate connections between supply and demand and promote high-level interaction between R&D and application sides, thereby forming a market-driven and internally sustainable agent industry ecosystem.
(1) Promoting Industrial Cooperation
34. Cultivating Open-Source Innovation
Guide domestic open-source AI communities to strengthen their focus on agents and promote compatibility between agents and open-source chips, operating systems, and foundation models. Encourage enterprises, universities, and research institutions to participate in open-source projects involving agent frameworks, interaction interfaces, and toolchains, thereby promoting interoperability and increasing international influence.
35. Building Industrial Collaboration Platforms
Leverage the role of industrial collaboration platforms such as agent ecosystem alliances and technology validation laboratories to coordinate upstream and downstream industry players in conducting R&D on common technologies, standards formulation, and evaluation and certification. Promote interdisciplinary talent development combining agent technologies and industrial applications. Encourage enterprises in internet applications and intelligent terminals to jointly build ecosystems and explore mutually beneficial cooperation models.
(2) Strengthening Application Promotion
36. Building Application Promotion Channels
Promote the establishment of agent software stores and industry supply-demand information platforms to encourage enterprises to release products and form agglomeration effects. Organize supply-demand matchmaking activities and use mechanisms such as public tenders and challenge-based competitions to attract enterprises to develop customized products. Encourage hardware and software companies to develop products and services based on agents and cultivate user markets.
37. Advancing the Opening of Key Scenarios
Promote the opening of key application scenarios in critical sectors and carry out pilot programs in industrial clusters, major industries, and priority areas to create leading demonstration projects. Develop market-oriented and professional technology transfer service institutions and explore agent deployment scenarios to improve the commercialization efficiency of technological achievements. Promote industry data sharing and openness to support the training and deployment of agents in key scenarios.
38. Actively Cultivating a Global Ecosystem
Leverage international platforms such as the World Artificial Intelligence Conference and the World Internet Conference to showcase and exchange innovations in agent technologies. Promote the adaptation of agents by terminal device and software enterprises, guide relevant enterprises in overseas compliance efforts, and support the localization of agents to comply with local laws, regulations, and cultural practices.
VI. Safeguard Measures
The Cyberspace Administration of China, the National Development and Reform Commission, and the Ministry of Industry and Information Technology, together with relevant authorities, shall strengthen overall planning and coordination, improve resource integration and collaborative efforts, refine supporting policies, and form a coordinated working mechanism to ensure implementation of key tasks. A comprehensive evaluation indicator system for agent development shall be established and improved, and monitoring, evaluation, rolling implementation, and dynamic adjustment mechanisms shall be strengthened to support the standardized application and innovative development of agents.
Q&A with the Press
Recently, the Cyberspace Administration of China, the National Development and Reform Commission, and the Ministry of Industry and Information Technology jointly issued the Implementation Opinions on the Standardized Application and Innovative Development of Agents (“Implementation Opinions”). A responsible official from the Cyberspace Administration of China answered questions from the press on the document.
1. Question: Could you introduce the background to the issuance of the Implementation Opinions?
Answer: In recent years, agent products represented by mobile assistants, terminal-based intelligent managers, and cloud-based agents have emerged rapidly and are being applied at scale, greatly facilitating people’s work and daily lives. At the same time, the high autonomy and high permission levels of agents have also created security risks such as privacy leakage, unauthorized operations, and loss of behavioral control. It is therefore necessary to coordinate development and security, and to promote the standardized application and innovative development of agents.
The CPC Central Committee and the State Council attach great importance to the development of artificial intelligence. General Secretary Xi Jinping emphasized at the 20th group study session of the Political Bureau of the CPC Central Committee that it is necessary to grasp the trends and laws of AI development, accelerate the formulation and improvement of relevant laws and regulations, policy systems, application norms, and ethical guidelines, and build systems for technological monitoring, risk warning, and emergency response, so as to ensure that AI is safe, reliable, and controllable. In August 2025, the State Council issued the Opinions on Deepening the Implementation of the “AI+” Initiative, focusing on areas such as science and technology, industrial development, consumption upgrading, public welfare, and governance capacity. It set a phased target that, by 2027, AI would be among the first to achieve broad and deep integration with key sectors, and the adoption rate of next-generation intelligent terminals, agents, and other applications would exceed 70%.
The formulation and issuance of the Implementation Opinions is a concrete measure to implement the Opinions on Deepening the Implementation of the “AI+” Initiative. Oriented toward promoting technological innovation, enhancing governance capacity, building industrial ecosystems, and improving public welfare, the document aims to create a favorable policy environment, leverage the demonstrative effect of typical application scenarios, and coordinate efforts to advance high-quality development, high-level security, and high-efficiency governance of agents.
2. Question: What is the overall thinking behind the Implementation Opinions?
Answer: In drafting the Implementation Opinions, emphasis was placed on four aspects. First, strengthening ideological guidance. The document adheres to Xi Jinping Thought on Socialism with Chinese Characteristics for a New Era, implements General Secretary Xi Jinping’s important thinking on building China into a cyber power, deepens the implementation of the “AI+” initiative, and promotes the high-quality development of agents. Second, coordinating development and security. It follows the objective laws of technological development, strengthens the industrial and technological foundation, improves the governance system, actively and prudently promotes the standardized application of agents, and builds a pattern in which development and security reinforce each other. Third, adhering to application-driven development. It deeply explores typical application scenarios, promotes a virtuous cycle in which innovation drives application and application promotes innovation, and enables agents to empower all sectors. Fourth, safeguarding the security baseline. It upholds a people-centered, AI-for-good, multi-stakeholder governance, and prudent security approach, clarifies the bottom lines and red lines for agent development, and embeds safety and controllability throughout the entire process of technology R&D, application deployment, and product promotion.
3. Question: What specific requirements does the Implementation Opinions set out for regulating the application of agents?
Answer: The Implementation Opinions takes the safety, reliability, and trustworthiness of agents as baseline requirements for industrial development, and promotes the orderly and standardized deployment of agents. First, it clarifies product principles. Policies, regulations, and ethical norms should be further improved, and work on permission management and behavioral control for agents should be strengthened to provide guidance for the R&D of related products. Second, it prevents security risks. Technological measures should be used to enhance risk prevention capabilities in areas such as intrinsic agent security, supply chain security, and application-derived risks, thereby achieving full-lifecycle security management across agent development, deployment, application, and maintenance, and effectively preventing systemic risks. Third, it improves the governance system. In line with the evolution of agent technologies, a prudent and sound tiered and classified governance framework should be established based on application scenarios and potential impacts. The construction of a compliance service system for agents should be promoted, so that agents can be both “allowed to develop dynamically” and “well regulated.” Fourth, it strengthens industry self-regulation. Agent-related enterprises, industry organizations, and research institutions are encouraged to enhance their sense of responsibility and jointly formulate industry self-regulatory rules. Developers, development platforms, distribution platforms, and service providers of agents are guided to participate in credit evaluation, jointly fostering a sound development environment.
4. Question: What key tasks does the Implementation Opinions lay out for promoting the innovative development of agents?
Answer: The Implementation Opinions systematically promotes the innovative development of agents around key directions such as technological breakthroughs, scenario-based applications, and ecosystem building. First, it strengthens the foundations for development. By reinforcing basic technology R&D and improving agent toolchains, it provides a high-level technological foundation for the industry. It establishes an agent standards system to lower the barriers to agent R&D, adaptation, and application. It also takes a forward-looking approach to frontier areas such as multi-agent collaboration and the intelligent internet, laying the groundwork for the industry’s continued evolution. Second, it strengthens application-driven development. Around areas such as scientific research, industrial development, consumption promotion, public welfare, and social governance, it proposes 19 typical application scenarios to drive the optimization of agent technologies and products and explore replicable and scalable deployment models. Third, it builds an innovation ecosystem. It strengthens the cultivation of open-source innovation forces and builds industrial collaboration platforms to promote efficient industrial coordination and enhance innovation vitality. By establishing application promotion channels and opening key scenarios, it improves supply-demand connectivity and forms a market-driven, internally motivated industrial ecosystem. It also actively cultivates a global ecosystem, promotes the integration and development of domestic and international technologies, and fosters an open and shared environment for international cooperation.
5. Question: What measures will be taken next to implement the Implementation Opinions?
Answer: The Cyberspace Administration of China, the National Development and Reform Commission, and the Ministry of Industry and Information Technology will work with relevant parties to effectively advance the implementation of the Implementation Opinions. Focusing on key areas such as agent technology R&D, scenario opening, and security governance, they will improve supporting policies, form coordinated working mechanisms, and promote the implementation of key tasks. At the same time, they will strengthen monitoring and evaluation, rolling implementation, and dynamic adjustment of the standardized application and innovative development of agents.