China does not yet have a dedicated AI statute. So when disputes over chatbots, deepfakes or robotaxis reach the courts, judges need guidance under existing law.
On 7 September 2026, China’s Supreme People’s Court (SPC) — the country’s highest court — held a press conference where Vice President Tao Kaiyuan released the Opinions on lawfully hearing AI-related dispute cases (最高人民法院关于依法审理涉人工智能纠纷案件的意见). The SPC calls it the first judicial-rules document on AI from the top court. It notes that China has no dedicated AI law, so the Opinions apply existing statutes [1].
The primary source for this draft is the SPC’s official Chinese release with the full text. Summaries below follow the research note’s translations. Law firm Jones Day’s English commentary was checked against that official text; where the firm glosses, it is labelled [1][2].
What kind of document is this?
These are Opinions — court guidance on how to hear cases — not a standalone AI Act passed by the legislature. The SPC says they apply existing statutes. The research note did not include a document number or same-day effective date from the official text [1].
The document has five parts and 24 articles [1].
Liability: fault by default
Article 3: fault-based liability under Civil Code article 1165(1), unless a law sets strict or presumed-fault liability. Courts weigh the setting, degree of autonomy, transparency, risk and preventive measures [1].
Article 2 sets principles: put people first, support innovation, hold a safety baseline. Liability should differ between general-purpose and specialised models, and between open and closed models [1].
Jones Day says strict liability applies to makers and sellers of hardware AI products. The official text says producers and sellers bear product liability under existing law; “strict” is the firm’s gloss [1][2].
Generative AI, personality rights and notice-and-takedown
Articles 4–5: AI face and voice cloning, “AI resurrection” of the dead, and AI-assisted doxxing can infringe personality and privacy rights [1].
Article 7: a generative AI provider is liable if it fails to act promptly after a valid notice, which needs preliminary evidence and the rights holder’s real identity. Users who maliciously prompt for infringing output are liable [1].
Article 6 (on personal data in training): for model training, processing personal information the person made public, or that was otherwise lawfully public, within a reasonable scope, and where the person has not expressly refused, is generally not infringement. Consent is needed where there is a major impact on the individual [1].
Physical products, driving and commerce
Article 9: physical AI products fall under the Product Quality Law. Defect factors include nature and purpose, self-learning, updates, user control and standards compliance, with a focus on warnings about limitations [1].
Article 10: algorithmic price discrimination, and punitive damages for fraudulent AI fake-celebrity endorsements [1].
Article 11: autonomous and assisted-driving crashes. Courts may order makers or operators to produce driving event data [1].
Training data, open source and inventorship
Article 12: a developer that claims non-infringement must produce evidence of training data sources, training records, how the model runs and its scientific basis. Users who knowingly generate substantially similar works are liable [1].
Article 13: open-source providers that release free code modules and disclose their functions and security risks may be found not liable for others’ infringement — a safe harbour of sorts [1].
Article 14: AI inventions can be patentable. A natural person who makes a creative contribution is the inventor — humans, not the model, get inventorship [1].
AI in the courtroom itself
Article 19: litigants who file AI-generated documents or case-search reports must verify them, disclose AI assistance and answer for their accuracy [1].
What the SPC left blank
The SPC says it deliberately left contested issues blank. AI-generated work copyrightability is not addressed [1].
What we still don’t know
- A document number and same-day effective date appeared only in a non-official search result and were not included in the research note.
- Jones Day’s commentary mis-states a 2021 Shaoxing case as violating an article of this 2026 document; treat that as an illustrative earlier case, not a literal citation [1].
- Jones Day’s summary leaves out Article 6 (personal data in training) and Article 19 (AI-generated court filings) [1].
The Bottom Line
China’s top court has published 24 articles of guidance for AI disputes: fault-based liability by default, notice-and-takedown for generative AI providers, product-liability factors for physical AI, a duty on developers claiming non-infringement to produce training-data evidence, an open-source safe harbour, human inventorship, and a duty on litigants to verify and disclose AI-generated filings. These are Opinions applying existing law — not a new AI statute — and the court itself left some hard questions, including copyright in AI-generated works, for another day.
Sources
- Supreme People’s Court of China, official release and full text, 7 September 2026 (Chinese). https://www.court.gov.cn/zixun/xiangqing/511101.html
- Jones Day, September 2026. https://www.jonesday.com/en/insights/2026/09/chinas-supreme-court-issues-landmark-judicial-rules-on-artificial-intelligenceimplications-for-product-liability-intelle
