On September 7, 2026, the Supreme People’s Court (SPC) of China issued its first-ever national judicial rules on artificial intelligence disputes — the Opinions on Lawfully Adjudicating AI-Related Dispute Cases (Fa Fa [2026] No. 10). The document does not create a new AI law; instead, it tells courts how to apply existing laws — the Civil Code, the Personal Information Protection Law, the Copyright Law, the Product Quality Law and others — to 24 concrete scenarios, from deepfakes to autonomous driving. For foreign companies operating in China, this is the clearest signal yet of where legal liability will fall when AI goes wrong.
- September 7, 2026: the SPC issued the Opinions (Fa Fa [2026] No. 10) — the first national judicial rules on AI disputes.
- 24 articles in 5 parts: covering tort liability, intellectual property, procedure and evidence, and working mechanisms.
- Fault-based liability is the default: strict liability or presumed fault applies only where a law expressly says so.
- AI products are physical-carrier products only: robots and autonomous vehicles qualify; pure AI services without a physical carrier do not.
- Deepfakes and voice cloning of an identifiable person require consent; unauthorized use can infringe personality rights.
- Publicly available personal information may generally be used for model training within a reasonable scope unless the individual expressly refuses.
Table of Contents
What Are China’s New AI Dispute Rules?
China has not yet enacted a standalone artificial intelligence law. The SPC Opinions fill part of that gap at the adjudication level: they guide how lower courts should resolve AI-related civil, intellectual property and criminal disputes under laws already in force. The full text is available in English on this site — see the Opinions of the Supreme People’s Court on Lawfully Adjudicating Artificial Intelligence-Related Dispute Cases — Full English Translation (2026) — and the official Chinese version, together with the drafting background and a Q&A session with SPC officials, is published on the Supreme People’s Court website.
The court’s stated approach is “development and security in parallel”: punish clear abuses — deepfake fraud, online doxxing, price discrimination — while keeping liability rules measured so that a young industry is not overburdened. The accompanying Q&A with responsible SPC officials makes that balancing explicit.
Why the SPC Opinions Matter to Foreign Companies
Foreign businesses touch AI in China every day: marketing teams generate images and video with AI tools, customer service runs on chatbots, HR departments use AI screening, and manufacturers ship smart devices. Each of those activities now has a clearer liability map. Three structural points deserve attention from any foreign-invested company operating in China:
- The rules apply through existing laws. There is no grace period and no grandfathering — the Opinions interpret the Civil Code, PIPL and other statutes that already bind your China entity.
- Liability is allocated across the AI chain. Courts will weigh the roles of developers, providers and users — which means your contract structure and internal controls determine how much risk you absorb.
- Users are not shielded by providers. A person who deliberately prompts an AI tool to generate infringing content bears liability directly, and the provider is pulled in only if it ignores a proper notice.
Key Liability Rules Your China Team Should Know
| Scenario | What the Rules Say |
|---|---|
| Deepfakes, AI face-swaps and voice cloning | Using a person’s name, likeness or voice to generate an identifiable digital likeness or synthesized voice without consent infringes personality rights; using it to spread false statements can infringe reputation. Consent and clear usage rights are essential in marketing and advertising. |
| Generative AI output that infringes others | A provider is liable for damage if it fails to stop generating infringing content after a valid notice with preliminary evidence and the right holder’s identity. Malicious prompters are liable themselves. |
| AI model training data | Processing publicly available personal information within a reasonable scope for training is generally not an infringement unless the individual expressly refuses; processing with major impact on individuals requires consent. |
| AI products | Only physical-carrier AI products (smart robots, autonomous vehicles) attract product liability. Producers and sellers must give truthful explanations and clear warnings about scenarios, limits and foreseeable risks. |
| AI-generated marketing content and copyright | A user who knows, or should know, of a prior work and generates a substantially similar work by AI without a reasonable defense is liable. Developers defending non-infringement must produce training data sources and process records. |
| Open-source AI components | Free open-source code modules with publicly disclosed functions and risks can shield their developers and providers from liability when others’ use causes infringement. |
The same part of the Opinions regulates consumer-facing practices: algorithm-driven differential pricing (“big data price discrimination”) can trigger liability where it materially restricts consumers’ rights to information, free choice and fair trade, and AI “celebrity impersonation” sales promotions that constitute fraud can lead to punitive damages. On the intellectual property side, AI-assisted inventions are patentable when a natural person made a creative contribution to the substantive features, and that person — not the machine — is the inventor.
A Compliance Checklist for AI Use in China
- Audit every AI touchpoint. List where AI is used in marketing, HR, customer service, product development and operations — then map each touchpoint to the relevant liability rule.
- Fix your content rights. Obtain and document consent or license rights before generating identifiable likenesses, voices or celebrity-style content; keep proof in your contract files.
- Review training data pipelines. Verify the legal basis for personal information used in training, track express refusals, and isolate high-impact processing for explicit consent.
- Build a notice-and-action process. If you operate a generative AI service, create a documented procedure for receiving right-holder notices and stopping infringing output promptly.
- Strengthen AI product documentation. For smart devices sold in China, align manuals, warnings and marketing claims with the actual automation level and known limits.
- Disclose AI in legal submissions. Any AI-generated court documents must be verified for accuracy and disclosed to the court — fabricated AI evidence can mean fines, detention or criminal liability.
For teams planning AI-driven business in Guangzhou or elsewhere in China, the right moment to run this checklist is before launch — and definitely before a dispute. Our bookkeeping and tax and legal services teams can help structure contracts, data processing records and product documentation so that liability stays with the party who should bear it.
Frequently Asked Questions
Do we need consent to use AI-generated images or voices of people in China marketing?
Yes. Generating an identifiable digital likeness from a person’s name or likeness, or a synthesized voice imitating a person’s timbre and pronunciation, without consent infringes personality rights under the new rules. Obtain documented consent or usage rights first — and never imitate celebrities to promote goods, which can constitute fraud with punitive damages.
Is our company liable if our AI tool generates infringing content?
A provider is liable for damage caused after it fails to promptly stop generating infringing content following a valid notice (preliminary evidence plus the right holder’s true identity). A user who deliberately prompts the AI to produce infringing content is liable directly. Both roles need documented procedures.
Can we train AI models on publicly available data from China?
Generally yes within a reasonable scope, where the individuals have not expressly refused. The court weighs purpose, data type, sensitivity and the context of publication. Processing with a major impact on individuals’ rights requires consent.
Do AI products face special liability rules in China?
Product liability applies only to AI in a physical carrier, such as smart robots and autonomous vehicles. Producers and sellers must truthfully describe applicable scenarios, inherent limits and foreseeable risks — misleading claims about automation levels create liability.
What happens if we use AI-generated documents in a Chinese court?
Litigation documents generated with AI must be verified for truthfulness and accuracy before submission, the use of AI assistance must be disclosed, and the submitting party bears legal responsibility. Using AI to fabricate evidence or stage false litigation can lead to dismissal, fines, detention or criminal liability.
China’s first national AI dispute rules reward companies that document consent, verify training data, test product claims and respond to notices quickly. If you want a practical compliance review of how your China operations use AI — from marketing content to smart products — contact Dan Young Business Consultancy. We will map your exposure against the SPC Opinions and the underlying statutes, and fix the gaps before anyone else finds them.
Disclaimer: This article is for informational purposes only and does not constitute legal, tax or professional advice. It summarizes judicial guidance issued on September 7, 2026, which interprets existing laws and is subject to further judicial practice and interpretation. Rules and enforcement practices may change. Please consult a qualified professional regarding your specific situation.
