Editor’s Note — DCC.
This is the question-and-answer session from the Supreme People’s Court (最高人民法院, SPC) press conference of 7 September 2026 at which the Court released the Opinions of the Supreme People’s Court on Trying Cases Involving Artificial Intelligence Disputes in Accordance with Law (关于依法审理涉人工智能纠纷案件的意见), document number Fa Fa [2026] No. 10 (法发〔2026〕10号). Vice President Tao Kaiyuan (陶凯元), Research Office director Zhou Jiahai (周加海), Civil Division No. 3 chief judge Li Jian (李剑) and Research Office deputy director Si Yanli (司艳丽) answered; spokesperson Ji Zhongbiao (姬忠彪) moderated. The Court’s transcript attributes the answers collectively to “the responsible officials of the relevant SPC departments”, and DCC keeps that attribution. The full text of the Opinions is on DCC’s law page; the Court’s release note, with DCC’s article-by-article map, is translated here. Where the first answer refers to “Vice President Tao’s earlier introduction”, it means that release presentation.
Five things in this transcript are worth more than the summary reporting has given them. First, the Court explains why it chose fault-based liability and a physical-carrier definition of “AI product”: not doctrine for its own sake, but a stated policy of not loading an early-stage industry with strict liability — the same conclusion the Hangzhou Internet Court reached in China’s first AI-hallucination judgment, now adopted from the top. Second, the “safe harbor” for generative AI is applied by analogy (参照适用) to Civil Code Article 1195, on the reasoning that a provider cannot pre-screen every output but can act on notice; under Article 7 of the Opinions that notice must include prima facie evidence and the rights holder’s real identity, and the exposure is for the harm that follows a failure to act. Third, on AI copyright the Court says liability “must match control capability and duty of care” and asks who “fed” the training data — and then places the evidentiary burden on the developer once the plaintiff shows generation and substantial similarity. Fourth, the Court is explicit that the two hardest questions — copyrightability of AI output and unlicensed training — were left out because the drafters could not agree. Fifth, the last answer confirms that Chinese courts have already encountered hallucinated case citations in filings, and that the Opinions’ duty to verify and disclose AI-generated pleadings is a response to that.
The translation is DCC’s. Chinese idioms the officials used for effect are kept and glossed rather than flattened.
On 7 September 2026, the Supreme People’s Court held a press conference on the Opinions of the Supreme People’s Court on Trying Cases Involving Artificial Intelligence Disputes in Accordance with Law. SPC Vice President Tao Kaiyuan, Research Office Director Zhou Jiahai, Civil Division No. 3 Chief Judge Li Jian and Research Office Deputy Director Si Yanli attended and answered reporters’ questions. The press conference was moderated by SPC spokesperson Ji Zhongbiao.
1. Rights protection versus innovation
Question: We note that the Opinions are grounded in current hot-button and difficult issues and respond in a targeted way to the concerns of the public and of the AI industry. In their specific content, how do the Opinions properly handle the relationship between protecting rights and interests and supporting innovation and development?
Answer: In his keynote speech at the opening of the 2026 World Artificial Intelligence Conference and High-Level Meeting on Global AI Governance, General Secretary Xi Jinping pointed out that we must always uphold equal emphasis on development and security, grasp the trends and laws of AI development, continuously improve the relevant laws and regulations, policies and systems, application norms and ethical guidelines, ensure that AI is safe, reliable and controllable, and let the thousand-li horse of AI run both fast and steady. The Opinions are built around “equal emphasis on development and security”: they regulate in accordance with law the abuse and malicious use of AI and strengthen judicial protection of people’s livelihood rights in the digital era, while unambiguously supporting technological and industrial innovation and escorting the healthy development of AI. Specifically:
On protecting people’s livelihood rights, the Opinions mainly provide as follows. First, they step up judicial protection of personality rights and interests. Unlawful conduct such as AI face-swapping, AI voice cloning and “online doxxing” (网络开盒) occurs from time to time and is a matter of great public concern. The Opinions therefore set clear rules for pursuing liability for using AI to infringe the rights to likeness, reputation, privacy and voice, so as to give full play to the normative and guiding function of adjudication and steer AI toward good. Second, they protect personal-information rights and interests in accordance with law. Large-scale, high-quality training data is the cornerstone of AI development. To balance protecting personal-information rights with promoting the AI industry, the Opinions — on the basis of the Civil Code, the Personal Information Protection Law and other laws — make clear that processing already lawfully public personal information within a reasonable scope for AI model training, where the individual has not expressly refused, is generally not found to be an infringement; but where there is a major impact on the individual’s rights and interests, consent shall be obtained as the law provides. Third, they regulate in accordance with law conduct such as “big-data price discrimination” (大数据杀熟) and “celebrity-impersonation selling” that infringes consumers’ lawful rights. Recently, algorithm-driven “big-data price discrimination” has been repeatedly exposed, and “celebrity-impersonation selling” has drawn close attention. We cannot expect every consumer to have “fiery eyes and golden pupils” (火眼金睛 — the Monkey King’s power to see through disguise); the law must draw its sword in time to protect consumers’ lawful rights. The Opinions therefore make clear that where, for the same goods or services, a business operator uses algorithms to impose unreasonable differential treatment in price or other transaction conditions, infringing others’ lawful rights and causing harm, the People’s Court shall find in accordance with law that it bears corresponding tort liability; and where AI is used for “celebrity-impersonation selling” that constitutes fraud and the consumer claims punitive damages, the People’s Court shall support the claim in accordance with law.
On supporting innovation and development, the Opinions mainly provide as follows. First, they accurately grasp the principle for attributing liability in AI-related torts, avoiding the imposition of excessive liability while AI technology and industry are still at an early stage of development, which would dampen the enthusiasm for innovation. The Opinions make clear that, except where the law expressly provides for no-fault liability or presumed-fault liability, legal liability for using AI to infringe civil rights and interests is governed by the fault principle. Second, they strictly define “AI product” in accordance with law to avoid the over-extension of product liability. The Opinions make clear that, strictly following the Product Quality Law’s definition of “product”, AI products are limited to products with a physical object as their carrier — intelligent robots and autonomous vehicles, for example — and AI services without a physical carrier are excluded. Third, they fully and accurately implement the criminal policy of combining leniency with severity in handling AI-related criminal cases. On the one hand, the Opinions stress that crimes committed by intentionally using AI to endanger national security, infringe citizens’ rights or disrupt social order are to be punished severely and resolutely in accordance with law; on the other hand, they require that innovative conduct in AI research, development and application be handled prudently in accordance with law, adhering to the principle of legality and the principle of proportionality between crime, responsibility and punishment. In addition, as Vice President Tao explained earlier, the Opinions make special provision for the proper trial of AI-related intellectual-property disputes, and those provisions likewise embody the idea of supporting innovation and development on the track of the rule of law.
2. AI-generated content and intellectual property
Question: New types of intellectual-property disputes arising from AI-generated content occur frequently and attract wide attention. What new provisions do the Opinions make on handling IP disputes involving AI-generated content?
Answer: Articles 12 to 16 of the Opinions are the provisions specifically addressing AI-related intellectual-property disputes, and Article 12 gives guidance on the AI-generated-content question you raise. The Outline of the 15th Five-Year Plan expressly proposes exploring the establishment of rules on the ownership of rights in AI-generated output and on the rights and responsibilities of developers, operators and users. The 15th Five-Year Plan for the Protection and Utilization of Intellectual Property recently issued by the State Council further stresses improving a copyright system adapted to AI development. The Opinions implement those central arrangements and respond promptly to public concern. Three points:
First, the Opinions clarify the factors for determining the liability of the relevant parties in cases where AI-generated content infringes copyright. AI-generated-content infringement involves multiple parties — AI technology developers, providers and users. Where AI-generated content is alleged to infringe another’s copyright, the relevant parties cannot be exempted from liability on the ground that the content was AI-generated; rather, the bearing of liability must match control capability and duty of care. In determining control capability and duty of care, one generally considers the technical and business-model characteristics of the AI at issue, the role each party played in the content-generation process, which party “fed” the relevant training data, the necessary measures taken, the profits obtained, and other factors. On that basis, whether the parties had subjective fault and to what degree is determined in accordance with law, and infringement liability is reasonably apportioned.
Second, the Opinions clarify the evidentiary rules for disputes in which AI-generated content infringes copyright. A rights holder claiming that an AI service developer or provider infringed its copyright must adduce evidence of the preliminary facts — that the infringing content was generated by that AI, and that it is substantially similar to the work in which rights are claimed. Because AI technology is complex and opaque, the relevant evidence is not something others can obtain. Therefore, where an AI developer raises a non-infringement defense, it must adduce evidence on the sources of the model’s training data, the training process, the mode of operation and the like, and where necessary provide the relevant scientific theoretical basis as corroboration. Where an AI user knows or ought to know of a prior work, uses AI to generate a work substantially similar to the prior work, and has no defense such as fair use, infringement shall be found.
Third, the Opinions regulate in accordance with law unfair competition carried out with AI-generated content. In the past two years, disputes over the use of AI to carry out unfair competition have been rising. Business operators use AI-generated text, images, videos and virtual characters to fabricate traffic and positive reviews, and even fabricate “popular-science” videos using AI face-swapping for false advertising. Such conduct not only harms others’ lawful rights and interests but seriously misleads consumers and disrupts the order of market competition; it shall be found in accordance with law to constitute passing-off infringement, false advertising and the like.
It should be explained that, during the drafting and argumentation of the Opinions, opinions diverged widely on the copyrightability of AI-generated content and on the characterization of using others’ works without permission to train AI large models, and understanding needs to deepen further. The Opinions therefore make no provision on these two questions for now. AI technology is developing at a breakneck pace, and new situations and new problems keep emerging. Next, we will continue to follow closely the intellectual-property protection issues in AI and other emerging fields, continuously understand and grasp the underlying laws in practice, continuously summarize adjudication experience, improve adjudication rules, and work to build consensus.
3. Personality rights: face-swapping, voice cloning, doxxing, injunctions
Question: As AI iterates faster, technologies such as deepfakes, intelligent data collection and user profiling have sharply lowered the threshold for infringement, and the protection of personality rights faces severe challenges. You mentioned that the Opinions address hot-button personality-rights issues such as “AI face-swapping and voice cloning” and “online doxxing”. Could you elaborate?
Answer: Personality rights bear on everyone’s personal dignity and are the most basic and important rights of civil subjects. We stress that the protection of personality rights and the development of the AI industry must be balanced with an inclusive and prudent attitude — but inclusiveness is not indulgence, and prudence is not laissez-faire. To respond to public concern and guide adjudication, the Opinions make targeted provision, by type, for the abuse and malicious use of AI technology that infringes personality rights. Mainly:
First, they regulate in accordance with law “AI face-swapping and voice cloning” and other uses of AI that infringe the rights to likeness, reputation and voice. In practice, the strongest complaints concern AI deepfakes. AI face-swapping can “harvest” anyone’s likeness indiscriminately, so that ordinary people, too, can become deepfake “targets”; with AI voice cloning, a few yuan will “steal a voice”, with results convincing enough to pass as real. Recently the news media also exposed a black-market chain selling “one-click AI undressing” software and tutorials, which the public detests. Article 4 of the Opinions provides expressly: using AI without consent to process a natural person’s name, likeness and the like to generate a virtual digital likeness identifiable as that person, and using or publishing it, infringes the right to name, the right to likeness and other personality rights and interests; using a natural person’s voice without consent as training corpus to imitate his or her timbre, intonation and pronunciation style and generate an identifiable synthetic voice infringes the voice rights and interests; manipulating a virtual likeness or synthetic voice to engage in improper conduct or make untrue statements that lower another’s social evaluation infringes the right to reputation — and the actor’s tort liability is pursued in accordance with law. In addition, on the phenomenon of “AI resurrection” of the deceased that has drawn public attention, the Opinions also make special provision to protect the personality interests of the deceased in accordance with law.
Second, they regulate in accordance with law the use of AI for “online doxxing”, “human-flesh search” (人肉搜索) and other conduct infringing the right to privacy. “Online doxxing” and “human-flesh search” are prominent forms of online violence; they seriously infringe victims’ right to privacy, seriously disturb the tranquility of their lives, undermine the public’s sense of security, and pollute the online ecosystem. We now see AI used to track and analyze a specific person’s public information — telephone numbers, online accounts, social media — in order to obtain private information, or to leak or publish the private information so obtained. Article 5 of the Opinions makes special provision on this.
Third, they clarify the rules for applying the personality-rights infringement injunction. Using AI to infringe personality rights is low-cost, spreads fast, has wide impact and causes harm that is hard to reverse; the traditional remedy of suing after the fact does not protect victims’ lawful rights promptly and effectively, so the lawful application of the personality-rights infringement injunction is especially important. Article 8 of the Opinions makes this clear. For example, a victim who has evidence that she is the target of AI face-swapped “sexual rumors” (黄谣), and that failing to stop it promptly would cause irreparable harm to her lawful rights, may apply to the People’s Court for a personality-rights infringement injunction. On review, where the conditions are met, the People’s Court may order the network user to cease the infringing conduct, and may also order the network service provider or AI service provider to take necessary measures promptly to prevent the harm from spreading further.
It should be stressed in particular that deepfakes and similar technologies are now used from time to time for fraud, insult, defamation, infringement of citizens’ personal information, and the production, sale and dissemination of obscene materials. Where such conduct constitutes a crime, it is to be severely punished and criminal liability pursued in accordance with law; the Opinions make special provision on this.
4. The “safe harbor” for generative AI
Question: Generative-AI software is now widely used, and the related civil disputes have drawn public attention. I note that, for determining infringement by generative AI, the Opinions introduce the “safe-harbor rule”. Could you go into that in depth?
Answer: Article 1195 of the Civil Code provides the “notice-and-takedown rule”, commonly called the “safe-harbor rule”. Simply put: when a network user uses a network service to commit an infringement (posting abuse about someone online, say), the rights holder, on discovering it, may notify the network service provider to take measures such as blocking or deletion; the provider, on receiving a qualified notice, must promptly take necessary measures, failing which it bears joint and several liability with the user for the expanded part of the harm. Conversely, if the provider promptly takes necessary measures after notice, it generally bears no tort liability.
In practice there has been controversy over whether generative-AI service providers may claim application of the “safe-harbor rule” by analogy. On study, we consider that although generative-AI services differ from traditional network services, the two are also similar in their operating mechanism. A generative-AI provider likewise organizes vast amounts of information and supplies it to network users; a large model’s output depends on the prior training data, the model parameters, the prompts the user enters and other factors; the provider can hardly foresee in advance everything users will input; and the volume of generated content is so enormous that a provider cannot be required to predict, review and intercept every output for infringement one by one. But once AI — through its own hallucination or a user’s malicious inducement — has generated infringing content, and the rights holder gives notice, the generative-AI provider should, and is in a position to, take necessary measures. There is therefore a legitimate and reasonable basis for applying the “safe-harbor rule” to generative-AI providers by analogy.
The Opinions accordingly provide for the determination of liability where “AI hallucination” infringes personality rights and where a user maliciously induces AI to do so. They make clear that: where content automatically generated by generative AI infringes personality rights and, after notice from the rights holder, the AI service provider fails to promptly take necessary measures such as ceasing to generate the infringing content, it bears tort liability in accordance with law for the harm caused; and where a network user maliciously induces generative AI to generate infringing content — by inputting infringing prompts or otherwise — and causes harm to another, that network user bears tort liability in accordance with law, and where, after notice from the rights holder, the AI service provider fails to promptly take necessary measures such as ceasing to generate the infringing content and blocking the relevant generation instructions, and the rights holder requests that the network user and the AI service provider bear civil liability, the People’s Court supports the request in accordance with law. These provisions accord with the laws of technological development and the legislative spirit of the Civil Code, and embody the organic unity of supporting innovation and governance in accordance with law; during the consultation process, all sides unanimously endorsed them.
5. AI in the courtroom: fabricated evidence, sham litigation, unverified filings
Question: AI’s impact is felt across every field, and adjudication itself faces new risks and challenges. For example, some litigation participants submit AI-generated evidence and other litigation materials to the court without verifying them first. How do the Opinions respond to such use of AI to obstruct the judicial order?
Answer: Litigation participants must observe the principle of good faith in litigation. As generative-AI technology has spread, it has become increasingly common for participants to use AI to assist their litigation. But AI is only an auxiliary tool; participants may not rely on “AI hallucination” or technology neutrality to escape their own legal responsibility. Participants must screen and verify the evidence and other litigation materials they submit to the court to ensure they are true and accurate. To guide the public in using AI reasonably, prevent false materials from entering proceedings, and effectively safeguard the judicial order, the Opinions make provision in three respects.
First, they punish in accordance with law sham litigation using AI. Where a party exploits AI’s self-learning and autonomous-decision-making characteristics to fabricate the facts of a case — by deleting or tampering with AI-generated and composed content labels, inputting specific instructions, selectively presenting results, adversarial interference or other human intervention or misdirection — and brings sham litigation, the People’s Court shall dismiss its claims in accordance with law and impose a fine or detention according to the seriousness of the circumstances; where a crime is constituted, criminal liability is pursued in accordance with law.
Second, they regulate in accordance with law the use of AI to fabricate evidence. The Opinions provide expressly that where a litigation participant or another person uses AI to fabricate evidence and obstructs the People’s Court’s trial of the case, the matter is handled under Article 114 of the Civil Procedure Law — that is, a fine or detention according to the seriousness of the circumstances, and, where a crime is constituted, criminal liability in accordance with law.
Third, they provide expressly that litigation participants bear a duty to verify and explain AI-generated litigation documents, case-search reports and the like. The People’s Courts have already found multiple instances of AI-generated false cases that the parties and their representatives did not fully verify before submitting to the court. In January this year the news media reported on such incidents in a concentrated way, and a case on the point has been entered into the People’s Court Case Database. On that basis, the Opinions provide expressly that where litigation documents, case-search reports or other materials submitted by a participant were generated using AI, the participant shall, before submission to the court, carefully verify the authenticity and accuracy of the laws, judicial interpretations, cases and other content involved, shall truthfully explain the use of AI assistance when submitting them, and shall bear responsibility in accordance with law for the authenticity and accuracy of the content. Courts abroad are likewise highly attentive to this problem; we note that some countries have issued dedicated guidance stressing that representatives are responsible for ensuring the accuracy of materials submitted to the court.
Source: 最高人民法院新闻局 (Supreme People’s Court Press Bureau), 《最高法相关部门负责人就〈关于依法审理涉人工智能纠纷案件的意见〉答记者问》, 最高人民法院 (Supreme People’s Court) WeChat Official Account, September 7, 2026. Original
— Not legal advice.