China's route to AI accountability: damages after the fact, not rules before it

On 7 September 2026 the Supreme People's Court of China presented its 'Opinion on the adjudication of artificial intelligence disputes in accordance with the law'. Chinese news agencies describe it as the first set of adjudication rules on AI issued by Beijing's highest judicial body: five parts, twenty-four articles. The move marks a clear methodological divergence from the European approach. The European Union — with the AI Act's transparency obligations in force since August 2026, and Italy at work on the implementing decrees of law 132/2025 — aims at regulating systems and providers before the fact. The Chinese document instead concentrates on redrawing civil liability, compensation for harm and procedural remedies after it.

The text takes a pragmatic run at several grey areas of the technology. Face swapping and voice cloning ('AI换脸拟声') without consent, used to create recognisable digital likenesses, are treated as an injury to personality rights. Service providers are liable if they fail to act promptly once notified; users are liable if they deliberately manipulate systems to generate harmful content. On the commercial side, the text establishes that a company using algorithms to charge different customers different prices for the same product, with no reasonable basis for the difference, answers for the harm caused to the buyers affected ('大数据杀熟'), and it lets consumers deceived by imitations of well-known figures claim punitive damages. It also covers protection against doxxing ('网络开盒') carried out by aggregating public data with AI, together with questions of autonomous driving and, on the intellectual property side, liability tied to open source software, the granting and confirmation of patents, the performance of technology contracts and the criteria for using data — plus a duty on the parties to check the filings and evidence they submit in court when these were produced with the help of AI.

At the press conference, Supreme Court vice-president Tao Kaiyuan explained that the aim of the act is to balance development and security, adding that points on which no consensus exists were deliberately left open pending further practical experience. Zhou Jiahai, head of the Court's research office, stressed that no one can expect every consumer to become an expert at spotting deception, which is what makes timely intervention by the law necessary. On method, though, one caveat is due: the full text of the opinion is not currently available through the official court.gov.cn channels. Our account rests on cross-checking the Xinhua dispatch, the China News Service report and the coverage in Jiemian News and Sina Finance. The date the measure takes effect, and its exact legal standing compared with a formal binding judicial interpretation, also remain unspecified.

Brussels and Beijing are not regulating the same thing. The AI Act places obligations on the provider before the system reaches the market; the Opinion gives judges the criteria for assigning liability and damages once the harm has occurred. They are different instruments, and this document does not tell us which of the two protects people better.

— Olya