On September 7, 2026, the Supreme People's Court issued the Opinions on the Lawful Hearing of Cases Involving Artificial Intelligence Disputes, Fa Fa [2026] No. 10. The Opinions are not a standalone “AI Law”: the SPC expressly notes that China has not enacted a dedicated artificial-intelligence statute. They are national judicial guidance for applying existing civil, intellectual-property, consumer, data and procedural law to AI disputes.
The 24 provisions address fault, personality rights, personal information, platform responsibility, AI-enabled products, automated driving, intellectual property, open-source software, technical contracts, evidence and litigation conduct. For businesses, the immediate task is to map each AI system, actor, affected right and evidence source to the applicable existing law and the new adjudication guidance.
Subject to editorial and legal review. This article is general information, not advice on a particular product, dispute or jurisdiction.
What the SPC Opinions do
The Opinions are the first AI-focused judicial adjudication rules issued by a national supreme court, according to the SPC. They guide Chinese courts rather than creating a comprehensive licensing code. The framework uses existing statutes, including the Civil Code, Cybersecurity Law, Data Security Law, Copyright Law, Anti-Unfair Competition Law, Consumer Rights Protection Law, Personal Information Protection Law and Civil Procedure Law.
Default fault-based liability
Courts ordinarily apply fault-based tort liability unless legislation provides strict liability or a presumption of fault. The assessment may account for the system's autonomy, transparency, foreseeable risks and the parties' ability to prevent harm. Businesses should therefore preserve design decisions, testing, risk assessments, warnings, deployment limits, monitoring and incident response.
Personality rights, deepfakes and doxxing
AI-generated images, avatars or voices can infringe portrait, name, voice, reputation, privacy or other personality rights when the person is identifiable. The Opinions also address the use of AI to aggregate public information to expose private details or disturb personal peace, and claims involving simulated representations of deceased persons. Courts may grant personality-rights injunctions where the statutory conditions are met.
Users, providers and notice-and-action
A user who deliberately prompts or uses an AI service to commit an infringement may bear direct responsibility. A generative-AI provider's exposure depends on its own duties and conduct, including knowledge and response to a valid rights notice. Providers need a documented notice channel, prompt and output controls, proportionate action, escalation, appeal and evidence retention; “safe harbor” should not be treated as automatic immunity.
AI products and physical systems
The Opinions connect product-liability analysis to whether the relevant AI product has a tangible physical form. Software embedded in robots, vehicles or other physical products may be assessed as part of that product, while a purely digital service does not automatically become a product for strict product-liability purposes. Contract, tort, consumer and sector-specific duties may still apply to software and services.
Public personal information and model training
Use of lawfully disclosed personal information for model training is not automatically unlawful, but processing must remain within a reasonable scope and respect an individual's clear objection where applicable. Separate consent or another legal basis may be required where processing materially affects individual rights. Teams should record provenance, purpose, scope, opt-out handling, sensitive information and downstream uses.
Intellectual property and deliberate open questions
The Opinions address AI-assisted infringement, open-source obligations, patent inventorship, technical contracts and data use. An AI system cannot be named as a patent inventor; a natural person must make the required creative contribution. The SPC deliberately leaves some disputed questions for further practice, including the legal status of outputs lacking sufficient human contribution and aspects of using copyrighted works in model training.
Evidence and AI-assisted litigation
Courts should scrutinize the authenticity and integrity of electronic evidence, including source data and analytical methods. A party using AI to fabricate evidence or facts may face dismissal, fines, detention or criminal referral where the legal conditions are met. AI-assisted pleadings and research must be verified before filing, disclosed as required, and remain the submitting participant's responsibility.
Business implementation priorities
- Inventory models, providers, deployers, users and physical products.
- Map affected personality, privacy, consumer, IP and contractual rights.
- Preserve training-data provenance, prompts, outputs, versions, notices and actions.
- Document testing, foreseeable misuse, warnings and human oversight.
- Build deepfake, doxxing, discriminatory-pricing and deceptive-marketing controls.
- Review open-source licences, technical contracts and patent contribution records.
- Adopt a verified and disclosed process for AI-assisted litigation work.
China, EU and US comparison
| Dimension | China | European Union | United States |
|---|---|---|---|
| Primary mechanism | SPC adjudication guidance applying existing laws to disputes | Risk-based AI Act obligations plus general product, data and IP law | Federal and state statutes, agency enforcement and case-by-case litigation |
| Deepfakes | Personality-rights and provider/user responsibility | AI Act marking and disclosure duties apply from August 2026, subject to scope and exceptions | Federal and state rules remain issue- and jurisdiction-specific |
| Products | AI product analysis focuses on tangible physical form | AI Act and revised product-liability framework can reach software-related risks | Product characterization and strict liability vary by governing law |
| Copyright | Human contribution and case-specific infringement analysis; some training and output questions remain open | AI Act transparency and copyright-policy duties interact with EU copyright law | Human authorship remains central; training disputes continue under copyright law |
The withdrawn proposal for an EU AI Liability Directive should not be described as operative law. Likewise, Section 230 treatment of generative-AI outputs in the United States should be treated as context-dependent rather than a settled universal rule.
Questions and answers
Is this a new comprehensive Chinese AI statute?
No. It is authoritative judicial guidance for applying existing law to AI disputes.
Is a provider automatically liable for harmful user output?
No. Liability depends on the governing rule, the provider's duties, knowledge, conduct and response, and the user's role.
Can an AI system be a patent inventor?
No. The Opinions require the relevant inventive contribution to be made by a natural person.
What must litigants do with AI-assisted materials?
Verify their authenticity and accuracy before filing, disclose AI assistance when required, and accept responsibility for the submission.
Official sources
- Supreme People's Court: Opinions on the Lawful Hearing of Cases Involving Artificial Intelligence Disputes
- Supreme People's Court: official questions and answers
- SPC Intellectual Property Court: understanding and application
- European Commission: AI Act framework and 2026 application timeline
- US Copyright Office: Copyright and Artificial Intelligence reports
Law checked: September 11, 2026. Confirm later SPC interpretations, guiding cases and applicable sector rules before acting.


