An Insights briefing from Muzy & Meraris LLP

By Muzamil Naeem

While the world debates how to regulate artificial intelligence — the EU with a sweeping statute, the United States with a patchwork of state laws and courtroom battles — China’s highest court has taken a different, quieter route. In September 2026, the Supreme People’s Court (SPC) issued the Opinions on Lawfully Adjudicating Disputes Involving Artificial Intelligence — its first comprehensive judicial guidance dedicated to AI. It is the first document of its kind from a national supreme court anywhere, and for any business that builds, deploys, or relies on AI in connection with China, it repays close attention.

The Opinions do not try to answer every hard question about AI. What they do is more subtle, and arguably more durable: they tell China’s courts how to think about AI disputes, and they do so around a single organising idea — that AI is a tool, and tools do not bear responsibility; people do.

What it is — and, importantly, what it is not

The first thing to understand is that the Opinions are not a new AI law, and not even a formal judicial interpretation. China, like most countries, has no unified “AI Act.” Rather than wait for one, the SPC issued a set of twenty-four provisions directing the courts to apply the laws China already has — the Civil Code, the Copyright Law, the Patent Law, the Personal Information Protection Law (PIPL), the Product Quality Law and others — to the novel situations AI creates.

In other words, the Opinions fill a gap not by inventing a separate body of “AI law,” but by mapping existing legal principles onto new technology. The core message, as commentators have put it, is that AI tools are to be treated like any other tool. That choice matters: it means AI is being folded into the ordinary legal order rather than carved out into a special regime.

The document is organised into five parts — general principles, civil infringement, intellectual-property disputes, procedural and criminal rules, and institutional mechanisms — and is built on three stated principles: putting people first (preventing the abuse of technology from harming personal dignity and rights), fostering innovation-driven development, and safeguarding a safety baseline. Development and security, the court says, are to be given equal weight.

Who is liable when AI causes harm

The Opinions’ most consequential contribution is on liability — the question everyone deploying AI eventually asks: if it goes wrong, who answers for it?

AI has no legal personhood. The Opinions state plainly that AI itself does not possess civil subject status. It cannot be a defendant, cannot hold rights, and cannot bear liability. Responsibility is borne by the civil subjects behind it — developers, service providers, and users. This disposes, for China, of the science-fiction notion that an autonomous system might somehow answer for itself.

Fault-based liability is the default. For AI-related torts, the Opinions make ordinary fault liability the general rule — you are liable if you were at fault — unless a specific law imposes something stricter (for example, strict liability under the Product Quality Law, or fault-presumption under the PIPL). Crucially, the guidance directs that responsibility should fall on those best positioned to control the risks and prevent the harm. That principle — liability follows control — cuts through much of the fog around AI accountability: the party with the power to prevent a harm is the party expected to answer for it.

Generative AI is not a “product.” In a notable classification, the Opinions indicate that pure, online generative-AI services do not fall within the scope of product liability. This has real consequences for how claims against AI providers are framed and which liability regime applies.

Intellectual property: lines drawn, and lines left open

The Opinions address several IP questions at the heart of the global AI debate — but with characteristic restraint.

On AI-assisted inventions, China has, for the first time, clarified that patent-eligible inventions still require an identifiable human inventor — aligning with the international consensus that an AI cannot be named as inventor. On AI-generated content, open-source software, technology contracts and data use, the Opinions provide a framework for analysis.

What they pointedly do not do is resolve the two most contested questions: the legality of large-scale training on copyrighted material, and the copyright status of AI-generated output. Those, the SPC signals, will continue to be decided case by case. This is a deliberate reticence — the court is drawing boundaries where it can and leaving genuinely unsettled questions to develop through litigation.

A firm line on personal data

On privacy, the Opinions take a clear position that businesses should note: they prohibit the use of personal information to train or optimise generative-AI models, reinforcing the protections of the PIPL and personality rights. For any company training models on data touching China, this is a direct and significant constraint — the kind of rule that must shape data-sourcing practices, not merely inform them.

AI in the courtroom itself

In a provision that resonates with headlines worldwide, the Opinions confront the problem of AI inside litigation — the “hallucinated” case citations and fabricated materials that have embarrassed lawyers and even judges in several countries. Parties who submit AI-generated materials to a Chinese court must verify their accuracy and disclose that AI was used. Courts, in turn, must rigorously examine AI-generated evidence, including the prompts used and whether results are reproducible. And fabricating evidence with AI can lead to adverse evidentiary treatment, fines, detention, or even criminal prosecution. The message to litigants and their lawyers is unambiguous: AI is a tool you remain fully responsible for.

The bigger picture — and why it matters beyond China

Step back, and the significance of the Opinions lies as much in the method as the content. The EU chose one comprehensive statute; the US has left much to a fragmented mix of state rules and individual court decisions. China has chosen a third path: direct the courts to apply existing law through authoritative guidance, and refine the rules over time through “typical cases.” It is incremental, pragmatic, and fast-moving — better suited, perhaps, to a technology that evolves faster than any legislature can amend a statute.

For businesses, the practical implications are concrete. Any company operating or deploying AI in connection with China should strengthen its governance across the full AI lifecycle — clarifying responsibilities among developers, providers and users in its contracts; tightening data-sourcing practices in light of the training-data prohibition; preserving the technical evidence (including logs and prompts) that AI litigation now demands; and ensuring that anyone using AI in a dispute verifies and discloses it.

And because this is the first comprehensive AI-dispute framework from a supreme court, it will be studied far beyond China’s borders, as other courts and legislatures grapple with the same questions. The Opinions are not the last word — even in China, the guiding cases to come will do much of the real work. But they are a landmark: the moment one of the world’s largest legal systems decided that artificial intelligence, for all its novelty, is a tool to be governed by the law we already have, wielded by the people who remain answerable for it.