On 7 September 2026, China's Supreme People's Court issued a set of 24 judicial provisions governing AI-related disputes, addressing unauthorised face-swapping and voice cloning, training data liability, algorithmic price discrimination and AI-generated misinformation – all within the existing framework of the Civil Code, Cybersecurity Law, Copyright Law and Personal Information Protection Law. The Opinions do not introduce new legislation; rather, they provide authoritative interpretive guidance on how courts should apply settled law to novel AI fact patterns. For international businesses operating in China, these provisions define enforceable standards that will shape litigation, regulatory exposure and contractual risk management from the date of their issuance onwards.
Context: Why Judicial Guidance Rather Than Legislation?
China's approach to AI governance has deliberately layered specialised administrative rules – the Algorithmic Recommendation Provisions, the Deep Synthesis Provisions and the Generative AI Measures – on top of horizontal statutes rather than enacting a single omnibus AI act. The Supreme People's Court Opinions of 7 September 2026 continue this pattern. By framing AI adjudication within existing law, the Court avoids a legislative gap while enabling rapid deployment of enforceable standards. The technique also gives courts interpretive flexibility: a judge may draw on personality-rights doctrine, data-protection obligations or intellectual-property principles depending on which cause of action is most apt on the specific facts.
This stands in contrast to the European Union's approach, where a dedicated regulation creates a tiered classification of high-risk AI systems and imposes ex ante conformity assessments. Clients who are already navigating AI Act compliance in European markets will find that the Chinese framework is structurally different: liability is fault-based, enforcement is predominantly ex post through litigation, and there is no formal list of prohibited or high-risk AI categories equivalent to the EU model. The two regimes can, however, interact significantly when a Chinese-developed AI tool is deployed in Europe, or when a European dataset is used to train a model commercialised in China.
The 24 Provisions: Structure and Scope
The Supreme People's Court Opinions are organised in five parts, covering: general principles of AI dispute adjudication; personality rights and biometric data; intellectual property and training data; market conduct and algorithmic fairness; and procedural and evidentiary matters. Each part applies existing statutory concepts to AI-specific scenarios rather than creating freestanding obligations.
- Part I – General principles: Courts must consider the degree of automation, the foreseeability of harm and the respective roles of AI developers, service providers and deployers when attributing liability. The multi-party chain common in AI deployment – foundation model developer, fine-tuner, API integrator, end-service operator – is explicitly acknowledged.
- Part II – Personality rights and biometric data: Unauthorised replication of a natural person's face or voice using AI techniques infringes personality rights under the Civil Code. Consent must be specific, informed and freely given; broad terms-of-service clauses that purport to authorise deep synthesis are unlikely to satisfy this standard.
- Part III – Intellectual property and training data: The Opinions address the contested question of whether scraping publicly available content for model training constitutes copyright infringement. The guidance applies existing fair-use principles and requires courts to assess purpose, market substitution and the degree to which the output reproduces protected expression.
- Part IV – Market conduct: Algorithmic price discrimination – offering materially different prices to similarly situated consumers based on profiling – is characterised as an unfair commercial practice subject to challenge under competition and consumer-protection law. AI-generated false information that causes reputational or economic harm attracts tort liability.
- Part V – Evidence and procedure: AI-generated content may be admitted as evidence subject to authentication requirements; the party relying on such content bears the burden of demonstrating its integrity and provenance.
Face-Swapping and Voice Cloning: Personality Rights in Practice
The treatment of deep synthesis technology is the most immediately actionable aspect of the Opinions for many businesses. Under the Civil Code's personality-rights chapter, every natural person holds rights over their likeness and voice. The Supreme People's Court has now confirmed that using AI to generate a realistic replica of someone's face or voice without authorisation – whether for commercial content, social-media posts or synthetic media – constitutes an infringement regardless of whether the underlying model was trained on the specific individual's data.
Practical consequences include the following. A brand that uses an AI-generated avatar resembling a real person in advertising without a valid licence risks a civil claim by that person, potentially aggregated with a regulatory referral to the Cyberspace Administration of China (CAC). An entertainment company that deploys a voice-cloned version of a performer – even a deceased one, given that personality rights can survive death for a defined period under Chinese law – must obtain rights from the appropriate right-holder. A technology provider offering a deep synthesis API must implement notice-and-action mechanisms: once notified that a specific replication infringes personality rights, prompt removal is required to avoid secondary liability.
The fault-based standard is significant. Demonstrating that a platform took reasonable technical precautions and acted promptly on notice can be a full or partial defence. This incentivises investment in detection and takedown infrastructure rather than strict precautionary prohibition.
Training Data: Copyright, Consent and the Open Questions
The training-data provisions are the most legally complex and, in some respects, the least settled. The Opinions apply existing copyright doctrine – specifically the analysis of whether a use is transformative, whether it replaces the market for the original work, and whether the output reproduces a substantial part of the protected expression – without establishing a bright-line rule that web-scraped training is either permitted or prohibited.
Several questions remain open after the Opinions:
- Whether a model that has merely learned from copyrighted text, without reproducing it verbatim in outputs, causes actionable infringement is not definitively resolved. Courts are directed to assess market substitution, which requires economic evidence about the impact on licensing markets for training data.
- The personal information dimension of training data is separately addressed under the Personal Information Protection Law. If training data includes personally identifiable information, a lawful processing basis – consent, legitimate interest or statutory authorisation – must be identified. The State Administration for Market Regulation (SAMR) and the CAC both have enforcement competence in this space.
- Ownership of AI-generated outputs remains subject to ongoing debate. The Opinions do not resolve whether a purely AI-generated work attracts copyright protection; they focus instead on liability for outputs that infringe third-party rights.
For international clients who have developed general-purpose AI models trained on global datasets, the training-data provisions create direct exposure in China if those models are deployed there commercially. The interaction with EU requirements – including the AI Office's guidance on GPAI model transparency and the copyright carve-out provisions of the AI Act – adds further complexity. We have examined the contractual structuring of GPAI deployment in detail in our analysis of GPAI contracts and liability in Germany, which provides a useful comparative reference.
Algorithmic Price Discrimination and AI-Generated Misinformation
Part IV of the Opinions targets two commercial practices that have attracted sustained regulatory scrutiny in China. Algorithmic price discrimination – where a platform charges a loyal or high-value customer more than a new customer for the same product, on the basis of profiling data – is addressed as an unfair trading practice. Courts are directed to examine whether the differential is justified by genuine cost differences or promotions; if not, it may give rise to claims under consumer-protection statutes and e-commerce regulations, and to referral to the SAMR.
Algorithmic transparency obligations embedded in earlier CAC rules are reinforced: operators of recommendation systems must disclose the principal criteria used to rank and price content or products. The Opinions confirm that courts may order disclosure of algorithmic parameters as part of an injunctive remedy, subject to trade-secret protections – a balance that will need to be struck case by case.
AI-generated false information is addressed under both defamation doctrine and the broader prohibition on spreading disinformation in China's information-management framework. An operator who deploys a generative AI tool that produces defamatory or false factual statements about an individual or company, and who fails to implement adequate safeguards or to act on notice, may face tort claims and administrative penalties concurrently.
Procedural and Evidentiary Dimensions
The evidentiary provisions in Part V are practically significant for dispute resolution. AI-generated content – whether a synthetic voice recording, a generated contract summary or a model-produced risk assessment – may be tendered as evidence, but the proponent must authenticate it: establishing the provenance of the underlying model, the integrity of the input data and the absence of post-generation manipulation. Blockchain timestamping and model-card documentation are increasingly relevant as authentication tools in this context.
For international clients considering dispute resolution strategies, the Opinions apply in proceedings before the Chinese courts. Arbitration seated in China – including before the China International Economic and Trade Arbitration Commission (CIETAC) – will be informed by these standards even where the tribunal applies foreign substantive law to contractual claims, because Chinese mandatory rules on personality rights and personal information protection apply regardless of choice-of-law clauses.
If your organisation is assessing how these procedural standards interact with your existing dispute resolution architecture, or considering whether to include AI-specific clauses in commercial contracts governed by Chinese law, our team is available to discuss your situation. Please contact us at info@ferrazwhitmore.com or visit our contacts page.
Cross-Border Implications for International Clients
The Opinions have extraterritorial relevance in two directions. First, a foreign AI developer whose product is used in China – whether through a local subsidiary, a distributor or direct cross-border access – is within the reach of the personality-rights and data-protection provisions if harm occurs in China. Second, a Chinese AI service provider seeking to expand into European or other markets must navigate frameworks that differ structurally from the Chinese one: the EU's classification of certain AI applications as high-risk AI systems, the prohibition on real-time biometric identification in public spaces, and the transparency requirements applicable to general-purpose AI models under the AI Act impose compliance obligations that have no direct equivalent in China's current framework.
There is also the employment dimension. AI tools used in Chinese workplaces – for performance monitoring, recruitment screening or salary benchmarking – engage personality-rights and personal-information obligations that the Opinions reinforce. Employers should review their AI-deployment policies in light of this guidance; our employment law practice in China advises on the full range of compliance considerations in this area.
For a comprehensive view of your organisation's AI-related legal exposure in China, including technology licensing, regulatory filings and litigation risk, our dedicated AI law practice in China provides integrated advisory support across these intersecting areas.
Strategic Recommendations
Based on the structure and content of the Supreme People's Court Opinions, international clients active in China should consider the following steps:
- Map your AI deployment chain: Identify every point at which your organisation develops, fine-tunes, integrates or deploys an AI model in China. Each node in the chain may attract distinct liability exposure under the fault-based framework.
- Audit consent mechanisms for biometric data: If your products process facial images, voice recordings or other biometric data, review whether existing consent forms satisfy the specificity and informed-consent standards now confirmed by the Court.
- Implement notice-and-action infrastructure: AI service providers that host user-generated or model-generated content should have functioning takedown procedures for personality-rights complaints, with documented response timelines.
- Review training-data provenance: For models deployed commercially in China, assess whether the training corpus included copyrighted material or personal information of Chinese data subjects, and whether appropriate licences or processing bases were established.
- Update dispute resolution clauses: Consider whether existing contracts include provisions addressing AI-generated evidence, data preservation obligations and the allocation of liability across the deployment chain.
- Coordinate China and EU compliance programmes: Where the same AI system operates in both jurisdictions, identify the points at which the Chinese fault-based regime and EU ex ante requirements interact or conflict, and design a harmonised compliance approach.
Outlook
The Supreme People's Court Opinions represent a considered and technically sophisticated response to the adjudication challenges posed by AI – one that prioritises interpretive coherence within existing law over the creation of a parallel statutory scheme. This approach is consistent with China's broader regulatory methodology in the AI sector, and it signals that the courts will continue to adapt established legal categories rather than await comprehensive legislation.
Several areas are likely to attract further judicial guidance or legislative action in the period ahead: the copyright status of AI-generated outputs, the precise scope of the training-data fair-use analysis, and the allocation of liability between foundation-model developers and downstream deployers remain contested. The interaction between Chinese AI governance and evolving international frameworks – including any future instruments developed under the auspices of multilateral bodies – will also require ongoing attention.
For organisations seeking to understand how the Supreme People's Court Opinions affect their specific commercial arrangements in China, or how they fit within a wider cross-border compliance programme, we welcome the opportunity to assist. Please reach out to our team at info@ferrazwhitmore.com or through our contacts page to arrange a discussion.
Disclaimer: This publication is provided for informational purposes only and does not constitute legal advice. The information herein should not be relied upon as a substitute for professional legal counsel tailored to your specific circumstances. Ferraz & Whitmore assumes no liability for actions taken or not taken based on the contents of this material. For advice regarding your particular situation, please contact info@ferrazwhitmore.com.