Picture a delegate at this year's Two Sessions (两会) standing up with a question that has dogged Chinese courts for years: when an autonomous system drafts a contract, misdiagnoses a patient, or floods a feed with synthetic propaganda, who is liable? The room had no single statute to point to.
China runs one of the world's largest deployments of artificial intelligence, yet it still has no unified Artificial Intelligence Law (人工智能法). What exists instead is a patchwork — a basic law here, a ministry rule there, and a much-discussed bill that, as of late 2026, exists only as research notes and expert drafts.
What "no AI law yet" actually means
The distinction matters because Chinese policy moves through clear stages, and confusing them produces bad reporting. The four stages are: the draft-for-comment (征求意见稿), the submitted draft (草案/送审稿), formal adoption and publication (正式通过/发布), and coming into force (生效实施). China's unified AI law has not reached any of the last three. It sits before the first.
- In the 2024 State Council legislative plan, a text described as the "Artificial Intelligence Law draft (人工智能法草案)" appeared as a preparatory item for submission to the Standing Committee of the National People's Congress (NPCSC) for deliberation.
- In the 2025 State Council plan, the wording was quietly downgraded to "advance the legislative work on the healthy development of artificial intelligence (推进人工智能健康发展立法工作)" — a shift from a concrete draft to an open-ended mandate.
- The NPCSC's 2025 and 2026 legislative work plans list AI-related legislation as a "preparatory review item (预备审议项目)," with the note that relevant bodies should "step up research and drafting and arrange deliberation as circumstances permit."
In plain terms: officials are studying and drafting, not voting. No public draft-for-comment has been issued, and no version has been submitted for a first reading. Anyone who writes that "China has passed an AI law" is describing a document that does not exist.
The committee work behind the silence
The delay is not inactivity. In 2025, five separate delegations — Zhejiang, Jilin, Liaoning, Shanghai, and Shaanxi — submitted proposals (议案) to the NPC calling for an AI law covering risk classification, regulatory mechanisms, and liability allocation. The NPC Education, Science, Culture and Health Committee reported that the Ministry of Science and Technology (科技部) and other departments are "steadily advancing" AI legislation, holding expert seminars, visiting AI firms for field research, and conducting "legal-text framework drafting (法律文本框架起草)."
That phrase is the tell. Framework drafting is earlier than drafting a full bill; it is the scaffolding, not the building. The 2026 plans do strengthen the language — the State Council's 2026 plan calls to "improve AI governance and accelerate comprehensive legislation for the healthy development of AI (加快推进人工智能健康发展综合性立法)" — but the NPCSC still files it under "preparatory review," not "under deliberation." Acceleration, yes; a bill, no.
The small-cut path that is already real
While the big law waits, China has been regulating AI through what scholars call small-cut legislation (小切口立法) — targeted rules for specific problems rather than one comprehensive statute.
- The Cybersecurity Law (网络安全法), first in force in 2017, was amended on 28 October 2025 and the revised text took effect on 1 January 2026. The amendment added a provision (Article 20) stating the state supports AI basic-theory research and key algorithm development, promotes training-data and compute (算力) infrastructure, improves AI ethics norms, and strengthens risk monitoring and security supervision.
- Earlier, standalone ministry rules already govern specific AI behaviors: the Provisions on Algorithm Recommendation (算法推荐规定), the Interim Measures for Generative AI Services (生成式人工智能服务管理暂行办法), and the Deep Synthesis Provisions (深度合成规定).
- The Measures for the Labeling of AI-Generated Content (生成式人工智能内容标识办法) took effect in 2025, requiring both generators and distribution platforms to mark synthetic material.
These are in force today. They are the real, operative AI governance in China — not the awaited comprehensive law.
Why Beijing is in no hurry
The hesitation is strategic, not accidental. A single statute written too early risks freezing a fast-moving field; written too late, it risks arriving after the practices it should govern have already hardened.
According to several legislative researchers, the chosen path is "act first where urgent, establish mature rules first, and improve dynamically (急用先行、成熟先立、动态完善)." Translation: fix the burning problems — algorithm abuse, synthetic disinformation, face-recognition overreach — with narrow rules now, and assemble the umbrella law only once patterns are clear.
Non-official expert drafts show the shape being debated. The Chinese Academy of Social Sciences published a model AI law (示范法) in August 2023, later revised to version 3.0 in 2025. In 2024, seven research institutions released a separate scholar draft (学者建议稿). Both are discussion documents, not bills. They signal direction — risk classification, transparency duties, liability allocation — without carrying legal force.
Honest limitations
This article describes the legislative research-and-drafting stage of the unified AI law as of late 2026; the final text, if and when adopted, may differ substantially. My account relies on publicly released legislative-work plans, NPC committee documents, and official CAC and NPC statements. It does not capture closed-door drafting language, nor the positions of individual agencies that may disagree behind the scenes. The exact contents of any future draft-for-comment remain unknown, and the timeline for adoption is political, not calendrical. Treat the "preparatory review item" status as a statement of current intent, not a promise of passage.
What readers can do now
- If you build or deploy AI in China, comply with the rules that are already in force — algorithm filing, generative-AI service measures, content labeling (内容标识), and the revised Cybersecurity Law — rather than waiting for a single umbrella statute.
- Track two official sources: the State Council legislative work plan (usually published each May) and the NPCSC plan, watching for the wording to move from "research and drafting" toward "draft (草案)" — that shift, not a press headline, is the real signal a law is coming.
- When you read "China's AI law," check whether the writer means the comprehensive bill (still absent) or the existing patchwork (very real). The difference changes what you must actually do.
