China has introduced its first nationwide regulatory framework for AI agents, combining risk-based oversight with measures aimed at accelerating adoption of the technology.
China unveiled its first national policy framework for artificial intelligence agents, introducing a tiered governance model based on application scenarios and potential risks, as Beijing seeks to balance rapid development of the technology with tighter regulatory oversight.
The document, jointly released by China’s internet regulator, top economic planner and industry authorities, describes AI agents as a potentially disruptive force capable of reshaping productivity, social governance and everyday life.
The framework underscores Beijing’s nuanced approach toward AI, encouraging innovation and industrial adoption while placing “safe, reliable and trustworthy” development as a non-negotiable bottom line. Authorities said AI-agent development should follow principles including safety and controllability, orderly regulation, innovation-driven growth and application-oriented deployment.
The document laid out measures in four broad areas: strengthening technological foundations and standards, reinforcing security and governance, expanding industrial applications and building an innovative ecosystem.
To strengthen technological foundations of AI agents, regulators called for improving the performance of general-purpose AI models, supporting specialized industry models and expanding access to high-quality datasets for training and deployment. The document also urges advances in capabilities of AI agents, including task planning, long-term memory, tool use and multi-agent collaboration, while accelerating development of testing, deployment, operations and security-monitoring tools.
The framework also proposes building an “intelligent internet” architecture for AI agents, including a registration platform offering digital-identity management, search and discovery as well as capability disclosure. Authorities called for stronger multi-agent coordination and research into technologies such as trusted connectivity, secure payments and conflict resolution, while leveraging IPv6 infrastructure to improve end-to-end communication and developing monitoring systems for the emerging network.
While highlighting the transformative potential of AI agents, officials warned that their autonomy and reliance on vast amounts of data create new security and governance risks, potentially rendering traditional cybersecurity models ineffective as the systems gain greater decision-making capabilities.
Tiered oversight, dual-defense systems
To address these risks, authorities proposed a tiered governance structure differentiating oversight according to risk levels and application scenarios. AI agents deployed in sensitive sectors and key industries will face stricter oversight, with cyberspace regulators and industry authorities determining approved deployment scenarios and imposing measures including filing requirements, testing obligations and potential product recalls.
Lower-risk applications such as entertainment and office-productivity tools will instead rely more heavily on compliance self-assessments, platform governance and industry self-regulation, reflecting Beijing’s effort to avoid overregulation and preserve corporate-innovation incentives.
The framework also proposes a dual-defense system combining “behavioral controls” with built-in security safeguards for AI agents. Regulators called for clearer boundaries between user-controlled actions, tasks requiring explicit authorization and autonomous decision-making, while ensuring users retain the right to be informed and have the final say over AI-driven actions.
Authorities emphasized technologies such as embedded rules and behavioral guardrails to keep AI agents operating within legal and ethical boundaries in public and private settings. The document also proposes using blockchain and related technologies to make AI-agent behavior in critical applications traceable and verifiable.
The framework further calls for stronger built-in security capabilities, including safeguards around data security, personal-information protection, encryption, attack detection and permission management, while proposing security evaluation systems for AI agents.
Authorities also pushed for broader standards covering key technologies, data exchange and trusted certification, alongside third-party compliance services including risk monitoring, testing and assessment, as well as certification service. The measures are aimed at lowering compliance costs through mutual recognition of certification results.
Applications, international governance
The framework identifies 19 application scenarios spanning scientific research, industrial development, consumer spending, public services and social governance as Beijing seeks to accelerate commercialization of AI-agent technologies.
In a nod to intensifying global competition in AI, the document calls on China’s open-source AI community to expand development of AI agents and improve their compatibility with open-source chips, operating systems and large language models, reducing reliance on foreign technology stacks.
China also pledged to participate more actively in international standards-setting and global AI governance through platforms such as the World Artificial Intelligence Conference and the World Internet Conference. Authorities said they would encourage device makers and software companies to adapt products for AI agents while guiding companies to strengthen foreign compliance efforts and ensure systems conform with local laws, regulations and cultural norms in foreign markets.
Source: MLex
