China has issued a non-binding five-year informatization plan that names AI filing, agents and AI legislation

Key takeaways

• The plan is a policy roadmap with 2030 targets. It states no penalty, no deadline and no effective date.
• On AI, it says to optimise the generative-AI filing mechanism and improve assessment and filing for new forms such as agents, on-device large models and anthropomorphic interaction services.
• It lists AI legislation and AI judicial interpretations as work to advance or study, without naming a bill or a date.
• For agents it points to standards work, including a maturity evaluation system and identity marking.

Is the plan binding, and who issued it?

It is a policy plan. The Q&A page says the Central Cyberspace Affairs Commission issued the National Informatization 15th Five-Year Plan and that the Office of the Commission prepared it with relevant departments. The Q&A calls it an action guide for informatization work in regions and departments, and says the Office will monitor implementation and publish a national informatization development report periodically.

The plan sets 2030 targets. The Q&A says there are 12 main indicators, and the indicators table in the PDF lists them against 2025 baselines. One of them is the adoption rate of next-generation smart terminals, agents and similar applications, which the table puts at 40 percent in 2025 and above 90 percent in 2030, classed as "expected" (预期性). Neither the plan nor the Q&A states a penalty, a filing deadline or an effective date.

The Q&A says the plan has 10 areas, 25 key tasks and 7 special columns. The PDF we opened numbers its special columns 1 to 8, so we do not repeat the count of seven as a fact about the document. Our Asia-Pacific AI regulation tracker collects the binding Chinese instruments separately, and the full list of briefings is on the AI regulation news hub.

What does the plan say about AI governance and filing?

The AI governance language sits in Part 10, on security support. It says to build an agile AI governance mechanism, optimise the generative-AI filing mechanism, set up a risk-assessment system for key application areas, and 完善智能体、端侧大模型、拟人化互动服务等新形态的评估备案 (improve assessment and filing for new forms such as agents, on-device large models and anthropomorphic interaction services). The same passage also mentions strengthening an AI cryptographic protection system.

The wording is directional. It says to optimise and improve; it does not say who files, by when, or on what standard. We cover the existing instruments in separate briefings: the agents opinions, the anthropomorphic interaction measures and the on-device filing item. The plan does not cite or amend them.

Elsewhere the plan lists work on agents without tying it to a duty. Part 2 says to promote the standardised application and innovative development of agents. Special column 4 says to formulate requirement guidance for key scenarios, build a technology maturity evaluation system for agents, and develop standards such as agent identity marking. Special column 5 says to explore AI social-impact assessment.

What does it say about AI legislation, judicial interpretations and standards?

Part 11 is the legal-support part. It says to advance foundational legislation on 推进人工智能、数字经济、电信、网络犯罪防治、反网络暴力等基础性立法。 (AI, the digital economy, telecommunications, cybercrime prevention and anti-cyberviolence). It names no bill, draft or timetable for any of them.

On courts, the plan says to 聚焦人工智能、个人信息保护、网络生态治理等热点问题，研究发布相关司法解释。 (focus on hot issues such as AI, personal information protection and network ecosystem governance, and study and publish related judicial interpretations). The text says "study and publish", which covers both steps; it gives no publication timetable.

Part 11 also says to speed up standards for AI, agents and smart terminals. Part 9 says to take part in international AI rules and standards. Part 10(2) says to promote and regulate cross-border data flows and to implement the data export security assessment, personal information export standard contract and certification systems. Special column 8 says to explore a cross-border data service centre.

What does it mean for counsel and AI product teams serving China?

For in-house counsel and compliance teams, the plan is a planning input, not an obligation. It names the categories for which it says to improve assessment and filing: agents, on-device large models and anthropomorphic interaction services. The plan does not say whether these are covered today. A product that combines an agent with on-device inference and a companion-style interface touches all three named categories.

For foreign providers, note that special column 7 is about Chinese enterprises going abroad: it says to explore overseas-business compliance guidance centres for key markets and to build a case library of legal disputes for those enterprises. It is not a service for foreign providers entering China. Teams advising lawyers can also see our attorneys and AI regulation hub.

What should teams do now?

• Inventory the China-facing features that are agents, run a model on the user's device, or present an anthropomorphic persona, as an internal planning exercise, and note which filings each already has.
• Keep the plan out of your compliance matrix as a rule. Record it as a signal, and tie each affected feature to the binding instrument that applies today.
• Watch CAC announcements on risk assessment of key application areas, agent standards and AI legislation, since the plan names these but sets no dates.
• If you rely on translations of the plan, check the Chinese text. Several lines say to improve or study where an English summary can read as a command or as new coverage.

What dates and numbers does the plan give?

The plan gives targets and an indicator, not deadlines for companies. The table lists the dates and figures it and the CAC Q&A state.

What we did not verify

What we opened: the 14-page plan PDF, the CAC publication page for the plan and the CAC question-and-answer page, all on cac.gov.cn, on 24 and 25 October 2026. Quotations are from the PDF text, which we extracted by machine, and the Chinese is given as it appears there.

What we did not open: the gov.cn reprint, any implementing rule, any CAC statement on filing for agents or on-device models, and any official English translation. The plan PDF carries no date line, so 23 October 2026 is only the date on the CAC pages, and we do not treat it as the issue date. The English glosses in brackets are our own translations.

What we refuse to claim: that the plan creates a filing duty, that agents or on-device models are or are not already covered by filing, that the plan enlarges any existing regime, that any legislation has a timetable, or what regulators will do beyond what the plan says.

Related briefings

• CAC, NDRC and MIIT opinions on AI agents
• China measures on anthropomorphic AI interaction services
• CAC anti-cyberviolence law draft and AI labelling duties
• CAC on-device generative AI filing