On June 3, China Legal Science Network released the “First Batch of High-Quality Judicial Cases Related to Artificial Intelligence.” The selection was organized by the Network and Information Law Research Office of the Institute of Law, Chinese Academy of Social Sciences. The ten cases cover typical scenarios such as generative AI services, AI companion products, synthetic voice and image, AI face-swapping infringement, and criminal liability for assisted driving, setting clear boundaries for compliance management for AI service providers.

---

Case 1: Liang vs. a technology company in a network infringement liability dispute

First instance court: Hangzhou Internet Court, (2025) Zhe 0192 Minchu 18143

1.1 Basic facts of the case

A certain generative artificial intelligence application is a general-purpose intelligent dialogue application for text generation and information queries developed and operated by the defendant AI company based on its self-developed large language model.

On June 29, 2025, when Liang entered prompts in a generative AI application asking about university application information, the AI generated inaccurate information about the university’s main campus and offered to compensate Liang 100,000 yuan if the generated content was incorrect, and suggested Liang file a lawsuit with the Hangzhou Internet Court. Later, Liang provided a university admissions brochure to generative AI, which then admitted that inaccurate information was generated during the conversation. Liang claimed that the AI product developed by a certain company provided him with incorrect information, misleading him and causing him harm, and promised compensation, requesting the court to order the AI company to compensate him 9,999 yuan for losses as a result.

1.2 Reasoning of the judgment The first-instance court held that generative AI applications do not have civil subject status and cannot independently make legally binding declarations of intent. The “commitments” generated cannot be regarded as expressions of intent from service providers under the circumstances involved and do not have legal effect on such expressions.

Tort disputes arising from generative AI services should be subject to the principle of fault liability, not product liability. In this case, plaintiff Liang claims purely economic loss and cannot determine the illegality or illegality of the act based on the infringement of the rights itself; rather, it must be determined based on whether the defendant AI company has breached the duty of care. Although the AI-generated information in question is inaccurate, it does not constitute legally prohibited toxic, harmful, or illegal information.

In this case, the defendant AI company fully fulfilled its obligation to prominently prompt service functions and to guarantee the reliability of generated content. The actions involved were not at fault and did not harm the rights and interests of plaintiff Liang, so it should be recognized as not constituting infringement according to law.

1.3 Reasons for selection

This case is a high-quality judicial case concerning the determination of liability for generative AI service providers. The judgment in this case clarifies:

1. Artificial intelligence does not have civil subject status and cannot make legally binding declarations of intent;

2. Generative AI provides services, not products, so fault-free product liability does not apply; fault-based liability should apply;

3. Generative AI service providers do not bear general review obligations for output content, but must strictly review toxic, harmful, or illegal information prohibited by law;

4. Generative AI service providers should fulfill the obligation to clearly prompt and explain service functions, so that the public understands the limitations of AI functions and serves as a warning and reminder;

5. Generative AI service providers must fulfill their basic obligations to guarantee functional reliability and adopt industry-standard technical measures to continuously improve the accuracy and reliability of generated content.

This case holds great reference value for future courts in China when trying similar cases.

---

Case 2: He vs. Shanghai Artificial Intelligence Technology Co., Ltd. Online Infringement Dispute

First instance court: Beijing Internet Court, (2020) Jing 0491 Minchu 9526

Court of Second Instance: Beijing No. 4 Intermediate People’s Court, (2021) Jing 04 Min Zhong No. 777

2.1 Basic Facts of the Case

The defendant, Shanghai Artificial Intelligence Technology Co., Ltd., operates an intelligent bookkeeping software. Users can create or add “AI companions” within the software, and set their name, avatar, relationship with the user, mutual address, and reply language. The software refers to some of these settings as “tuning.”

The plaintiff, He, is a public figure. Without He’s authorization, an “AI companion” using He’s name and likeness appeared in the software involved. This character is open to other users, who can upload portrait images of He’s “memes” and other materials, create graphic and text materials, and interact and “train” the “AI companion” through related features.

2.2 Reasoning of the Judgment

The appellate court held that creating an interactive and “trainable” AI virtual avatar based on a natural person’s name, likeness, and personality traits essentially recognizes, simulates, and utilizes the natural person’s personality image. Network service providers who organize users to participate in content production through product features, software rules, and algorithm applications, and realize core service functions and commercial interests through relevant content, do not qualify as purely neutral technical service providers. If a natural person’s name, likeness, or other personality marks are used without the rights holder’s permission, resulting in damage to related personality rights and interests, tort liability shall be borne according to law.

2.3 Reasons for Selection

This case is a high-quality judicial case concerning AI companion products. AI companionship products are created and interact externally by users, while platforms deeply participate in the creation, dissemination, and use of personalized virtual avatars through feature settings, interaction rules, and algorithms. People’s courts can penetrate the surface form of users uploading content themselves and passive storage by platforms, and combining product design, open mechanisms, and platform benefit methods to determine the responsibility of network service providers. This helps prevent AI companion products from evading personality rights protection obligations under the guise of user-generated content, and promotes the standardized development of intelligent interaction products based on respect for personal dignity and freedom.

---

Case 3: Shanghai Chuanghua Company sued Hangzhou parent company over copyright infringement and unfair competition dispute

First instance court: Hangzhou Internet Court, (2024) Zhe 0192 Minchu No. 1587

Second instance court: Hangzhou Intermediate People’s Court, (2024) Zhe 01 Min Zhong 10332

3.1 Basic Case Facts

A Shanghai company named Chuanghua has obtained the intellectual property and related rights protection rights for the Ultraman series images through authorization. The defendant is the operating entity of an AI platform. The platform calls the Stable Diffusion open-source model and integrates interface design, technology integration, data management, and application deployment according to platform needs and usage scenarios, completing a series of engineering operations such as encapsulation, transmission, and storage, ultimately integrating them into a generative AI platform that users can directly apply. The platform provides foundational models like Checkpoint and LoRA, supporting image-to-image and online model training. On the platform’s homepage and under “Recommended” and “IP Works,” there are intelligently generated Ultraman images and Ultraman LoRA models, which can be applied, downloaded, published, or shared through links. The Ultraman LoRA model is generated by users uploading Ultraman images, selecting the platform’s foundational model Checkpoint, adjusting parameters for training. Afterwards, other users can input prompts, select the base model Checkpoint, and overlay the Ultraman LoRA model to generate images substantially similar to Ultraman’s image.

3.2 Grounds for judgment

The first-instance court held that when a service provider provides generative AI technology services, whether it constitutes aiding infringement should comprehensively consider factors such as the provider’s profit model, the reputation and influence of the rights work, the prominence of the infringement facts, the level of AI technology development, the feasibility and cost of alternative designs to avoid damage, necessary measures and their effects that can be taken, and the impact of infringement liability on the industry, dynamically adjusting the criteria for determining fault. Limit the duty of care of service providers to a reasonable extent commensurate with their information management capabilities. Generative AI services are only subject to anti-unfair competition regulations if they violate the principles of good faith and recognized business ethics, disrupt market competition order, and harm the legitimate rights and interests of other operators or consumers.

3.3 Reasons for Selection

This case is a high-quality judicial case concerning copyright infringement and unfair competition involving generative AI platforms. The judgment in this case insists on balancing innovation promotion with law-based governance, safeguarding rights and supporting the development of the service industry in parallel, properly balancing the protection of creators’ legitimate rights and interests with promoting AI technological innovation. It provides important guidance and demonstration for the healthy and standardized development of the generative artificial intelligence industry. This case proposes practical and referable adjudication approaches, providing important guidance for judicial practice, corporate compliance, and industry development, and also provides a reference for courts nationwide in adjudicating copyright infringement and unfair competition disputes involving generative artificial intelligence. The trial and judgment of this case are not limited to resolving individual disputes, but also extend to the artificial intelligence industry, cultural industry, and the general public, conveying the concept of the rule of law for intellectual property protection and holding significant social value for creating a fair competitive market environment and promoting high-quality development of the digital cultural industry.

---

Case 4: Beijing Doumou Technology Co., Ltd. v. Yimoumou Information Technology (Beijing) Co., Ltd. Unfair Competition Dispute

First instance court: People’s Court of Chaoyang District, Beijing, (2023) Jing 0105 Minchu 71391

Court of Second Instance: Beijing Intellectual Property Court, (2023) Jing 73 Min Zhong 3802

4.1 Basic Case Facts

In June 2020, Beijing Doumou Technology Co., Ltd. launched AI-powered comic transformation effects on its “Doumou” app. This effect uses model structure and parameters to instantly convert user-generated content into a comic-style style corresponding to real human scale, and has attracted significant market attention since launch.

In August of the same year, Yimou Information Technology (Beijing) Co., Ltd. launched shoujo manga effects on its app. This effect closely matches the imaging visual effects of Douyin’s AI transformation comic, with model structures and parameters closely related to the two companies.

4.2 Reasoning of the Judgment

The appellate court held that the AI model structure and parameters formed by operators through data training and optimization tuning, condensing R&D investment, can bring innovation advantages and operating profits, and constitute legitimate rights protected by the Anti-Unfair Competition Law. If others commercialize and use relevant model structures and parameters without permission, thereby disrupting the order of fair competition, it constitutes unfair competition.

4.3 Reasons for Selection

This case is a high-quality judicial case concerning the structure and parameter competition law protection of AI models. When there are blind spots in the protection of copyright, trade secrets, and other legal paths, people’s courts have regulated unauthorized use of others’ model results under the Anti-Unfair Competition Law, reflecting the judiciary’s prudent expansion of protection methods for new technological achievements. This case helps protect investment in AI model R&D, curb free-riding, and maintain fair competition in the AI industry.

---

Case 5: Yin Mouzhen v. Beijing Mouwen Intelligent Technology Company and others regarding personality rights disputes

First instance court: Beijing Internet Court, (2023) Jing 0491 Minchu No. 12142

5.1 Basic facts of the case

The plaintiff, Yin Mouzhen, is a voice actor. In May 2023, Yin Mouzhen discovered that works created by others using his dubbing were being distributed across multiple apps. After sound screening and traceability, the voices in the related works come from text-to-speech products operated by Beijing Wen Intelligent Technology Company’s platform.

In 2019, Yin Mouzhen was commissioned by Zhongmou Audio (Beijing) Cultural Media Company to record audio recordings. Later, Zhongmou Audio & Video (Beijing) Cultural Media Company signed a data authorization agreement with Mouruan (China) Co., Ltd., providing the recording data, including the involved audio products, to Mouruan (China) Co., Ltd. Certain Software (China) Co., Ltd. used a recording made by Yin Mouzhen as material, processed it using artificial intelligence and other technologies to generate the text-to-speech product involved in the case, and sold it externally on the cloud service platform operated by a Shanghai cloud network technology company.

A Beijing Wen Intelligent Technology Company obtained the text-to-speech product involved in the case through an online service sales contract and accessed and used it on its platform via an application programming interface. The AI voice involved in the case is highly consistent with Yin Mouzhen’s timbre, intonation, and pronunciation style, with the platform showing the voice has been played over 3.2 billion times.

5.2 Reasoning of the Judgment

The first-instance court held that if the voice synthesized by AI technology is sufficient to enable the general public or the public in related fields to identify a specific natural person in terms of timbre, intonation, and pronunciation style, it still falls within the scope of voice rights protection. Copyright or data usage authorization for sound recordings does not necessarily include authorization for AI-generated modeling, generation, and commercial use of natural human voices. Processing and using identifiable natural human voices with artificial intelligence without the individual’s consent constitutes infringement of voice rights.

5.3 Reasons for selection

This case is a high-quality judicial case concerning the protection of personality rights in AI-generated voices. People’s courts regard identifiability as the core criterion for determining whether synthetic sounds are protected, distinguishing between the property rights of sound recordings, authorization to use voice data, and the personality rights of natural persons, and preventing data authorization chains from replacing rights holders’ autonomous decisions regarding voice personality interests. This case has strong practical relevance to protecting voice actors, announcers, and other subjects whose voice is their professional identity, and also helps guide the speech synthesis industry to establish mechanisms for legal authorization, prudent use, and traceable management.

---

Case 6: “AI Face-Swapping” Civil Public Interest Litigation on Personal Information Protection

First instance court: Hangzhou Intermediate People’s Court, (2025) Zhe 01 Minchu 2028

6.1 Basic facts of the case

Defendants Hu Moujia and Hu Mouyi obtained others’ personal information such as phone numbers, profile photos, and ID information through channels, and used AI face-swapping technology to create fake liveness authentication videos. The two also replaced local phone cameras by embedding virtual camera apps into their phones.

When the platform calls the camera for facial recognition verification, the system plays a pre-made authentication video to pass the platform’s verification process. The two individuals used the above methods to log into others’ accounts, obtain delivery addresses, shopping cart information, and other data, sell them for profit, and also provided paid fake real-name authentication services.

6.2 Reasoning of the Judgment

The first-instance court held that biometric information such as facial recognition constitutes sensitive personal information. Using AI face-swapping technology to create fake liveness authentication videos, bypassing facial recognition authentication mechanisms, and thereby illegally obtaining and selling others’ personal information and providing paid fake real-name authentication services infringes on the personal information rights of an unspecified majority and harms public interests, and should bear liability for infringement according to law.

6.3 Reasons for Selection

This case is a high-quality judicial case in public interest litigation related to AI-related personal information protection. People’s courts determined tort liability from three perspectives: the method of technology abuse, the scope of the victims, and the consequences of public interest damage, revealing the real risks posed to biometric authentication order by combining AI face-swapping with virtual cameras. The judgment in this case helps strengthen the protection of sensitive personal information, promotes platforms to improve liveness detection, security verification, and anomaly identification mechanisms, and also provides valuable reference for civil public interest litigation in personal information protection to address new technological risks.

---

Case 7: Sun sued Chengdu Tu Catering Management Co., Ltd. and Xu over disputes over portrait, voice, and name rights

First instance court: Chengdu Railway Transport First Court, (2025) Chuan 7101 Minchu 8546

7.1 Basic facts of the case

In February 2024, the defendant, Chengdu Tu Moumou Catering Management Co., Ltd., posted a video via its Douyin account. The video content features an AI-generated image and voice of Sun XX, with the synthesized character presenting New Year greetings to viewers as Sun XX and representing the company to extend New Year wishes to clients.

The video in question contains Sun’s name, likeness, and voice features, and also involves promoting the company’s brand. The video used highly recognizable AI-generated images and voices without Sun’s authorization.

7.2 Reasoning of the Judgment

The first-instance court held that voices and images processed by AI technology, as long as they have elements sufficient to identify a specific natural person, such as timbre, intonation, pronunciation style, and appearance, are still protected by personality rights. Combining synthetic voices or images with natural persons’ names without permission for corporate brand promotion or commercial promotion can easily lead the public to develop associations such as endorsements or recommendations, constituting infringement of name rights, portrait rights, and voice rights.

7.3 Reasons for selection

This case is a high-quality judicial case concerning AI-generated commercial marketing of audio and video. People’s courts evaluate synthetic voices, synthetic images, names, likenesses, and brand promotion content in combination, clearly stating that operators cannot circumvent personality rights authorization requirements on the grounds of synthetic technology processing. This case sets the boundaries of rights for enterprises using artificial intelligence to synthesize the images of celebrities, experts, and public figures for marketing activities, helping to reduce risks of impersonation, false associations, and the commercialization of personal interests.

---

Case 8: Wang Mouqun’s dangerous driving case

First instance court: People’s Court of Linping District, Hangzhou, (2025) Zhe 0113 Criminal First Instance No. 596

8.1 Basic Case Facts

From about 0:31 to 0:38 on September 13, 2025, defendant Wang Mouqun drove a small passenger car after drinking and returned from Jiakai City, Tangqi Town, Linping District, Hangzhou to Shui’an Jincheng Community. From about 1:15 to 1:18 a.m. the same day, defendant Wang Mouqun drove the vehicle out of the community, then continued driving on the road from the passenger seat using the car’s assistance system. During this period, defendant Wang Mouqun, in order to evade the system’s steering wheel detection, installed an accessory called the “intelligent driving miracle” on the steering wheel, causing the system to mistakenly believe the driver’s hands were still on the wheel. At around 1:37 a.m. on the same day, the vehicle stopped driving while passing the Chongyun Street section of Yingbin Road in Linping District, Hangzhou. After a report from the public to the police, the defendant Wang Mouqun was apprehended by the police. Tests showed that defendant Wang Mouqun’s blood ethanol content was 114.5mg/100ml.

8.2 Reasoning of the Judgment

The first-instance court held that at this stage, assisted driving systems are not equivalent to fully autonomous driving, and drivers still have the obligation to drive safely during vehicle operation. If the perpetrator activates the assisted driving function after being intoxicated and evades steering wheel detection by adding relevant accessories, causing the motor vehicle to drive on the road, it still constitutes drunk driving and constitutes the crime of dangerous driving.

8.3 Reasons for Selection

This case is a high-quality judicial case concerning improper use of assisted driving functions and criminal liability. The People’s Court, based on the current technical functional positioning and the driver’s actual control responsibility, rejected the defense of equating assisted driving with autonomous driving, and clarified that a drunk person operating the vehicle using the assisted driving system should still bear primary driving responsibility. This case helps correct the public’s excessive trust in intelligent driving functions, reminds drivers not to use technical devices to evade safety supervision, and promotes the standardized application of intelligent connected vehicle technology within the baseline of road traffic safety.

---

Case 9: Li v. Xuzhou Xinmou Cultural Media Co., Ltd. Online Tort Liability Dispute

First instance court: Beijing Internet Court, (2024) Jing 0491 Minchu No. 10085

9.1 Basic facts of the case

The plaintiff, Li XX, has a certain level of recognition and social influence in the fields of education and child-rearing. In 2024, the defendant Xuzhou Xinmou Cultural Media Co., Ltd. used Li Moumou’s public speech and teaching videos in a certain online platform store, accompanied by AI-synthesized voices highly similar to Li Moumou’s voice, to promote multiple family education books sold in the store.

The promotional content in question combined Li’s likeness and voice with related book sales activities, linking Li’s personal image and professional background to the defendant’s commercial promotional targets. The related videos were published by livestream sales hosts and used to promote the defendant’s store book sales.

9.2 Reasoning of the Judgment

The first-instance court held that using the likeness of a specific natural person and an AI-generated voice highly identical to their voice for online product promotion without permission is sufficient to cause the public to perceive a recommendation, endorsement, or other commercial cooperation relationship between the rights holder and the product seller, constituting infringement of portrait rights and voice rights. The entrusting party has a reasonable obligation to review the promotional content published by the promoter. If failure to do so causes infringement consequences, the entrusting party shall bear corresponding responsibility.

9.3 Reasons for selection

This case is a high-quality judicial case concerning the use of AI-generated voices for the protection of personality rights in online e-commerce. While confirming the rules for the identifiability protection of synthesized voices, the people’s courts further incorporated the review obligation in entrusted promotion relationships into liability determination, preventing merchants from evading responsibility by outsourcing sales or streamers publishing content. This case helps regulate the use of AI-generated voices in short videos and live-streaming e-commerce, promoting product promotion activities based on authenticity, legality, and respect for personal rights and interests.

---

Case 10: An international investment company sued Cao over online infringement liability dispute

First instance court: Guangzhou Internet Court, (2025) Yue 0192 Minchu No. 23599

10.1 Basic Case Facts

The plaintiff, an international investment company, is a listed domestic enterprise, and a senior executive of the company was arrested on suspicion of duty-related crimes. In August 2025, Cao entered a document of about 1,700 words into a generative AI application, detailing the executive’s embezzlement case and containing unverified information such as related-party transactions involving an international investment company.

Cao then sent instructions to the AI to generate a lengthy in-depth article of no less than 10,000 words. After multiple rounds of processing and online searching, the AI generated an article of over 15,000 words. The article introduced the personal experience of a certain executive and stated that the international investment company had abnormal financial data, inflated profits, benefit transfers in related-party transactions, and suspected financial fraud. Upon verification, there was no relevant content in the AI-cited sources. Cao published the article on his financial public account, marked as “Original,” with over 10,000 reads and more than 1,000 shares.

10.2 Reasoning for the Judgment

The first-instance court held that users, as the actual controllers of generative AI prompt input, content generation, and external dissemination, are the initiators and managers of infringement risks and cannot be exempted from liability on the grounds that the content is AI-generated. When users disseminate content related to others’ reputations generated by generative AI, necessary verification of the authenticity, objectivity, and reliability of the content shall be conducted, and appropriate labeling shall be made according to the nature of the content; If failing to exercise duty of care leads to false information infringing on the company’s reputation, the company shall bear tort liability according to law.

10.3 Reasons for Selection

This case is a high-quality judicial case concerning online infringement of AI-generated synthetic content. The People’s Court did not simply attribute responsibility to technical tools, but instead focused on the input of materials, generation instructions, online search, article publication, and dissemination effects, detailing and confirming the obligations of users and disseminators. This case helps curb the use of generative AI to create and amplify unverified information, guides content creators to exercise necessary caution when it comes to AI-generated information, and safeguards corporate reputation, rights, and online information order.

---

How should AI companies and developers comply?

Combined with the ten AI judicial cases released this time, the compliance boundaries between AI operating companies and technology developers are clearly defined, helping to avoid various infringement litigation risks.
1. Key points for AI enterprise operational compliance

Complete basic qualifications: For large models, AI face-swapping, and virtual human products, algorithm filing and security assessment must be completed; Adding manual verification to financial and medical AI cannot rely solely on models to generate professional conclusions.

Standardizing training data sources: commercial models are prohibited from scraping unauthorized text, text, and audio training sessions; Using others’ likenesses or vocal training requires separate written authorization. Bundled default authorization does not have legal effect, and a material authorization ledger is kept simultaneously.

Product risk control: prominent AI features artificial hallucinations on the page, automatically adds non-erasable watermarks to deeply synthesized content, and intercepts uploads of infringing materials; A fast complaint channel has been launched, handling infringement clues 24 hours a day, and user operation logs are retained long-term for evidence collection.

Uphold the bottom line of intellectual property: Prohibit copying competing model structures and parameters, complete internal compliance reviews before launch, and avoid unfair competition and copyright disputes.

2. AI Developer Development Considerations

Data source compliance: Reject biometric information such as copyright-free or undesensitized faces, voices, and other biometric data as training materials; annotate data to complete privacy de-identification.

Strictly adhere to R&D red lines: No reverse dismantling or copying of competitors’ commercial AI models, and no development of illegal tools such as AI watermark removal or fake facial authentication.

Built-in security mechanisms in the code: The program embeds a risk alert module to intercept AI-generated false claims such as compensation and guarantees, preventing misleading users.

Standardizing open source and delivery: Using open-source frameworks and datasets, strictly adhering to commercial agreements, and preserving development documentation completely; When delivering models, clearly inform partners of their material compliance responsibilities to avoid joint infringement risks.

---

Conclusion

The era of AI rule of law has officially arrived; technological freedom does not mean laws without boundaries. These ten benchmark cases set clear action standards for the entire industry, with clear rights and responsibilities for every scenario—from large model development, AI content creation, commercial marketing to intelligent driving—each scenario has clear rights and responsibilities. In the future, as AI-related disputes continue to increase, judicial authorities will keep releasing typical cases, build a complete and stable judicial protection system for artificial intelligence, and promote the healthy and orderly development of digital technology on the track of the rule of law.

[Statement] This article is compiled from typical judicial cases in the AI field from 2025 and is for readers’ research and reference only, and does not constitute legal advice.

Source: http://iolaw.cssn.cn/gg/ggqt/202606/t20260603_6009787.shtml

---

 

About Us

MeritsIP: Your Global Intellectual Property Partner

At MeritsIP, we provide comprehensive IP services in biomedicine, medical devices, manufacturing, semiconductors, and AI. We support 200+ top clients globally with strategic, high-quality IP services.

Stay Connected

• • Website: https://www.meritsipglobal.com

• • LinkedIn: Follow us on LinkedIn

• • Email: meritsip@meritsandtree.com

© 2025 MeritsIP. All Rights Reserved.