7/24 Service · Chinese / English

China AI Liability and Algorithmic Harm: Defending Foreign Companies in Tort and Regulatory Claims

China has no standalone Artificial Intelligence Law, so an AI-harm claim is defended under ordinary rules: general fault, network infringement, and strict product liability — together with the generative-AI and labelling duties that determine where fault is found.

China AI Liability and Algorithmic Harm: Defending Foreign Companies in Tort and Regulatory Claims

There is no standalone AI-liability law — and that is the first thing to know

As of late 2026, China has no dedicated statute allocating liability for harm caused by artificial intelligence. A draft Artificial Intelligence Law has been listed as a preparatory item in the legislative plan but has not been enacted, so the current position is best stated as "subject to the latest official progress". What exists instead is an assembled framework: the general tort rules of the Civil Code, the network-infringement provisions, the product-liability regime, and a set of sectoral regulations that shape — but do not by themselves create — liability. A foreign company defending an AI or algorithmic claim is therefore defending under ordinary rules, not a special AI statute.

The tort framework is assembled from the Civil Code

Three groups of Civil Code provisions do the work. Article 1165 supplies the general fault rule: a person who, through fault, infringes another's civil-law rights and causes damage bears liability. For AI-generated content that is unlawful, the network provisions apply — Articles 1194 to 1197 set up the notice-and-takedown structure, and Article 1197 makes a network service provider that knew or should have known of infringing content, and failed to act, jointly liable. Where the harm comes from a defective AI product or system, Article 1202 applies product liability, which is strict liability on the producer. There is no statutory fault presumption for AI harm — the oft-repeated claim that "AI harm attracts presumed fault" is not the current law, and a defence memo should not assert it.

The regulatory duties that shape liability

The sectoral rules matter because a breach of them is how fault is established. The Interim Measures for the Management of Generative AI Services (seven regulators, effective 15 August 2023) require training data to have a lawful source and not to infringe IP or personal-information rights (Article 7), make the provider bear content-producer responsibility (Article 9), require labelling of generated content (Article 12), and require prompt removal of unlawful content and reporting (Article 14). The Provisions on the Management of Algorithmic Recommendations (effective 1 March 2022) require prominent marking of algorithmically generated or synthesised content (Article 9), transparency about the algorithm's basic mechanism (Articles 16–17), and prohibit unreasonable differential treatment such as algorithmic price discrimination (Article 21). The newest layer is the Measures for the Labelling of AI-Generated and Synthesised Content (effective 1 September 2025), which mandates visible labels on generated content and invisible metadata in the file, and puts verification duties on the distribution platform. On the data side, the Personal Information Protection Law's Article 24 governs automated decision-making and requires an option to refuse decisions made solely by automated means where they materially affect personal rights.

How to defend an AI or algorithmic claim

  1. Classify the harm first: content harm is a network-infringement or fault question under Articles 1194–1197; a defective system or product is a strict-liability question under Article 1202; a discrimination or automated-decision complaint is primarily a PIPL Article 24 question.
  2. Fix the label and the notice posture: visible labelling and prompt takedown under the generative-AI and labelling rules are the two cheapest ways to deny fault.
  3. Keep the training and compliance record: a documented data-source and filtering process under Article 7 of the generative-AI measures is the evidence that converts a claimed breach into a defensible one.
  4. Do not assume a fault presumption: allocate the burden correctly — the claimant must prove fault in a general tort or network claim, and the main strict-liability exposure is the product-liability route, not AI-in-general.

— Danny Luo, Attorney-at-law, Tianni Law Firm, Nanjing. This article is provided for general information only and does not create an attorney-client relationship.

This article is provided by Tianni Law Firm for general information only and does not create an attorney-client relationship. For legal advice on a specific matter, please contact a qualified attorney.

Defend an AI or algorithmic claim in China.

We advise foreign companies on AI and algorithmic liability, content-harm and product-defect claims, and regulatory investigations by the cyberspace and market regulators.

Related Articles