China’s new AI agent framework puts healthcare in the regulatory spotlight

China has issued its first dedicated policy framework for intelligent agents, placing healthcare among the highest-priority sectors for both deployment and oversight as regulators move to govern a technology they describe as fundamentally reshaping production, daily life, and social governance.

The Cyberspace Administration of China (CAC), the National Development and Reform Commission (NDRC), and the Ministry of Industry and Information Technology (MIIT) jointly published the Implementation Opinions on the Standardised Application and Innovative Development of Intelligent Agents on 8 May 2026.

The document is the first time Chinese regulators have addressed intelligent agents as a distinct category — separate from generative AI and large language models — and sets out a comprehensive framework covering technology standards, safety requirements, governance structures, sector-specific applications, and ecosystem development.

The significance of the document lies not only in what it regulates but in what it signals. Previous Chinese AI policy focused primarily on generative AI, content moderation, model filing requirements, and data compliance. This framework treats agentic AI as an emerging layer of digital infrastructure in its own right — systems capable of autonomous perception, memory, decision-making, interaction, and execution, operating across both cyberspace and the physical world.

Regulators explicitly describe intelligent agents as “profoundly transforming modes of production, daily life, and social governance,” language that positions the technology as a foundational national priority rather than a discrete product category.

Healthcare as a priority and a test case

Healthcare receives specific and detailed attention throughout the document, and the provisions point toward a sector that regulators expect to see transformed by agentic AI in the near term.

On the clinical side, the framework calls for improvements to medical AI agents supporting imaging analysis, disease diagnosis, and the generation of personalised treatment plans. Drug management, surgical scheduling, and medical records management are all identified as active deployment targets.

For patients, the document highlights pre-consultation agents and report interpretation tools as near-term priorities — applications designed to reduce friction at the point of care and improve overall experience without requiring direct clinical involvement.

Crucially, healthcare falls within the document’s highest-risk governance tier. Any agent application in the sector will be subject to mandatory filing with regulators, compliance testing, and product recall provisions, administered jointly by internet regulators and sector-specific health authorities.

This places health AI under some of the most stringent oversight in the entire framework — alongside transportation, public security, and financial services — reflecting both the opportunity regulators see in the sector and the risks they are unwilling to leave ungoverned.

The combination of deployment ambition and strict oversight is deliberate. The document’s underlying logic is that China does not intend to slow the integration of AI into healthcare, but will require that integration to occur within a defined compliance architecture.

For health technology companies operating in or entering the Chinese market, the practical implication is that agent-based products in clinical or patient-facing contexts will need to be built with regulatory engagement as a first-order consideration, not an afterthought.

A framework built around tiered risk

Beyond healthcare, the document establishes a governance model applicable across 19 designated application sectors. The central mechanism is a tiered classification system based on application scenario and potential societal impact. Sensitive and high-priority sectors — including healthcare, finance, transportation, judicial services, and public security — face the most demanding requirements: mandatory filing, structured testing, and recall mechanisms operated jointly by the CAC and the relevant sectoral authority.

Lower-risk domains, including general consumer applications and office productivity tools, will be governed through a lighter-touch combination of self-assessment, information reporting, platform management, and industry self-regulation.

The document frames this as achieving both “vitality” and “control” — releasing enough regulatory flexibility to allow commercial development while preserving the state’s capacity to intervene when necessary.

The framework sets explicit boundaries around decision-making authority, a provision that has direct implications for product design across all sectors. It distinguishes between actions that must remain under direct human control, actions that may be delegated through user authorisation, and actions an agent may take autonomously.

Users must retain the right to be informed of any autonomous action and the right to override it. Agents may not act beyond the scope of a user’s explicit authorisation — a requirement that will shape how healthcare agents, financial agents, and government-facing agents are architected and marketed.

Additional safety provisions prohibit agents from exploiting personalisation or human-like interaction techniques to create emotional dependency, addictive behaviour, or manipulative consumption patterns. The document singles out minors and elderly users as groups requiring particular protection — a provision with clear relevance to consumer health and mental health applications.

Standards, infrastructure, and the intelligent internet

The document devotes considerable attention to the underlying infrastructure it expects to support a large-scale agent ecosystem. It calls for the development of an Agent Interoperability Protocol (AIP) as a key national and industry standard, and proposes the construction of an “Intelligent Internet” architecture — a term that signals ambitions well beyond regulating individual products.

The Intelligent Internet concept encompasses agent registration platforms, digital identity systems for agents, capability declaration mechanisms, trusted multi-agent coordination protocols, and compliant payment frameworks. IPv6 is identified as a foundational infrastructure layer for agent-to-agent communication.

Taken together, these provisions amount to early-stage planning for a future network layer in which AI systems interact with each other directly, with China seeking to define the protocols on which that interaction is built.

The document also calls for the development of open-source agent frameworks compatible with domestic chips and operating systems, consistent with China’s broader strategy of reducing dependence on foreign technology stacks across critical digital infrastructure.

China’s international positioning

The framework explicitly states China’s intention to participate actively in international standards-setting for intelligent agents. This is not an incidental provision. Global competition around agent standards is already under way, with organisations including the Linux Foundation having launched initiatives focused on agentic AI interoperability.

China’s explicit signalling that it intends to shape rather than adopt emerging global protocols reflects an understanding that the governance architecture for agentic AI — covering identity, permissions, payments, and accountability — will be as consequential as the technology itself.

For regional observers, particularly those tracking AI governance across Asia, the document represents a significant reference point. Several ASEAN member states are at early stages of developing their own frameworks for health AI and emerging AI applications.

China’s move to establish a comprehensive, sector-differentiated governance model for agentic AI — with healthcare at the centre — will inevitably influence how regulators and policymakers across the region approach the same questions.

Implementation

The CAC, NDRC, and MIIT are designated as joint implementation authorities. The document commits the three agencies to coordinating across government, refining supporting policies, and establishing monitoring and evaluation mechanisms that allow for rolling adjustment as the technology and its applications evolve.

The full text is published on the CAC’s official website at cac.gov.cn.