A recently publicised defamation case has moved the question of who bears responsibility for AI-generated content from the developer’s side to the user’s side.
The facts were broadly these: the defendant used a generative AI tool to produce a script, then made and published a video about an institution and an individual that fabricated facts and damaged their reputation. When sued, the defendant argued that “the video copy was written by AI and its content had an online basis”.
The court did not accept the defence. The core reasoning: users of generative AI services bear a duty to verify the relevant information. The defendant neither cited sources nor checked the accuracy and credibility of the material, and could not produce effective evidence for the content. The defendant therefore infringed the plaintiff’s reputation rights and bore corresponding liability.
Why the duty falls on the user
This is a different question from whether a developer is liable when AI errs.
On the developer’s side, the rule established by the case cited in the Supreme People’s Court work report is that where the developer has exercised due care and no rights were actually harmed, there is no infringement — see our note on the Supreme People’s Court work report on AI.
On the user’s side the logic is entirely different. The user decides whether to publish and has full control over that decision, and therefore bears a duty to verify. Treating the generation tool as a defence amounts to transferring publishing responsibility to a system that has no capacity to be a civil-law subject — which does not hold legally.
Three groups to note
Companies and brands. Where marketing and operations teams produce and publish content using AI, any inaccuracy is the company’s liability. A pre-publication verification process is needed.
Content platforms and MCNs. Where account networks use AI-generated content at scale, the verification duty falls on the operating entity. Platform-side review duties are discussed in our note on the AI face-swap drama case.
Publishers of corporate information. AI-generated false information about companies is now within the scope of targeted governance — see our note on the Qinglang campaign against malicious speculation about enterprises.
Four practical steps
First, cite sources. For content containing factual statements, citing the source is the lowest-cost compliance action and the key to evidencing diligence.
Second, tier the verification. For content touching on reputation, commercial evaluation or public events, raise the level and require at least one traceable primary source.
Third, keep records. Verification steps, revisions and approval chains should be traceable so that diligence can be shown in a dispute.
Fourth, bring AI use into internal policy. Specify where AI may be used, where human review is mandatory, and who is responsible for publication. The synthetic-content labelling duty must also be engineered — see our note on the CapCut enforcement case.
“Written by AI” is a statement of fact, not a defence. For companies, AI raises content-production efficiency but does not lower the publisher’s liability threshold.
This article is based on publicly reported judicial cases and related explanations from the Ministry of Justice’s legal-education platform. It is for general reference only and does not constitute legal advice.