China

Case management

The ‘Judicial Knowledge Services Platform’ is a judicial services platform jointly promoted by the Supreme People’s Court in collaboration with leading Chinese AI intelligence enterprises, including iFLYTEK, Hanvon, Alibaba DAMO Academy, Hikvision, and Taiji, to integrate advanced image, text, speech, and video technologies into court operations and to provide courts across China with a unified suite of AI-enabled services. It supports functions including digitisation and OCR of case files, automatic classification and indexing of court records, intelligent assistance at the case-filing stage to detect jurisdictional and procedural issues, recommendation of similar cases, and real-time speech recognition during hearings. In 2023, the platform reportedly managed 1.14 billion knowledge points and provided courts nationwide with 7.8 billion integrated intelligent services of the types described above.

Legal research, analysis and drafting support 

The ‘Intelligent Auxiliary System of Criminal Case Handling’ (206 System) (also known as the ‘Shanghai Criminal Case Intelligent Assistance System’), developed by the Shanghai High People’s Court and iFlyTek, is an AI-assisted system that supports judges in criminal trials by analysing case facts, identifying relevant legal issues, and recommending applicable laws and sentencing guidelines. It is now in its fourth iteration. The 206 System aims to enhance consistency and efficiency in criminal adjudication. 

During the development of the 206 System, more than 400 officials were assigned from the courts, the procuratorate, and the police to advise approximately 300 iFlyTek staff on the legal standards that should inform the computer code and the functionality of the software. The project initially focused on standardising outcomes in criminal cases. One mechanism for doing so was introducing automated checks to make sure each required piece of evidence was submitted. Official media reports on the launch of the 206 System noted that it was designed to address the three major causes of incorrectly-decided criminal cases: weak or illegal evidence, insufficient examination of evidence, and differences among judicial personnel handling criminal cases. The same source quoted a Shanghai police officer stating that the software could help reduce defects and omissions in the evidence-collection process. The 206 System also learns from past cases to give sentencing predictions and recommendations. The 206 System pulls from criminal case documents data about statutory punishments, benchmark punishment, and declared punishment. It also extracts information about sentencing circumstances, discretionary factors, and historical factors to create a training set for machine learning.

Analogous criminal case management systems designed by iFLYTEK have been adopted by judicial institutions in Anhui, Shanxi, Guizhou, Xinjiang, Shenzhen, Henan, Qinghai, and other provinces. But not all courts in the country are equivalently equipped.

Other courts have developed their own AI systems to assist with legal research, analysis and drafting: 

Decision-making support

AI has been used in China’s criminal justice system as early as 2006, when the Zichuan District People’s Court in Zibo, Shandong Province, introduced computer-based sentencing software co-developed with a high-tech company, marking the first attempt to apply AI-like technology in sentencing. Since then, several other local courts have developed their own sentencing assistance systems. 

Projects in places as diverse as Shanghai, Hainan, and Guangzhou are introducing software capable of analysing past cases with similar fact patterns to recommend sentences to judges. In a drunk-driving case, for example, a judge would select a list of factors (such as blood alcohol level or amount of damages caused), and the software would display the average sentence in past ‘similar’ cases. The judge retains discretion to disregard the recommended sentence.

One example, used in the Qingyuan Qincheng District Procuratorate in Guangdong Province since 2022, is ‘Little Judge Bao’, which predicts sentences based on legislative interpretations, judicial interpretations, and precedent cases. Little Judge Bao is now in service in various other courts and is also consulted by other political-legal organs. 

The Little Judge Bao Intelligent Sentencing Prediction System, discussed above, is also used by judges to suggest penalties based on big data analysis of case information and prior judgments from similar cases. Judges have liberty to ignore or reject the suggestions for criminal punishments. While the system appears to be operational in the criminal law context, details on specific jurisdictions or case types remain limited. 

The 206 System, discussed above, also generates sentencing recommendations for the reference of judges and prosecutors.

Risk-assessment

The Xiao’an Guardian (小安守卫) is a psychological assessment system jointly developed by the Mentougou District People’s Court of Beijing and Peking University, designed to assist judicial authorities in assessing the psychological state of minors involved in judicial proceedings. The system adopts a multimodal, dynamic assessment approach, supporting on-site interviews, text input, and audio recognition, and conducts structured linguistic analysis in units of approximately 550 characters. It generates indicators of psychological stability and emotional authenticity, categorises risks such as anxiety, depression, and self-harm into graded levels, and applies a four-tier classification to factors including negative psychological states and rationalisation patterns. Assessment reports generated by the system are subject to manual review and confirmation by judges and qualified psychological counsellors to determine whether further professional counselling or intervention is required.

Defence

Administrative support

The National Judicial AI Platform, discussed above, will provide public legal services after the platform is further trained and optimised. It will, for example, answer non-professional queries on legal issues, allowing individuals to access legal services and consultations more easily. 

The 12368 Intelligent Voice System (12368智能语音系统) is an AI application deployed under the national court system’s judicial information service hotline, with the participation of technical partners such as DeepSeek. Leveraging speech recognition, semantic understanding, and knowledge-base question-answering technologies, the system operates primarily as an ‘intelligent customer service’ platform, providing litigants with services including case status inquiries, complaints and reports, submission of opinions and suggestions, contact with judges, court information searches, procedural consultations, litigation service password inquiries, psychological support, and litigation-related petition services. Upon receiving a caller’s request, the system conducts semantic analysis and generates real-time responses on a round-the-clock basis, while allowing seamless transfer to human operators where automated handling is insufficient. As at March 2026, AI functionalities have been deployed in only a limited number of regions, principally Shanghai, Zhejiang, Jiangsu, and Anhui.

Legal research, analysis and drafting support 

‘Fa Xiaotao’ (法小淘) is an AI software that can assist lawyers in the preliminary search and analysis of a case. Based on the description of the facts of the case, it can analyse and calculate the proportion of winning or losing for similar cases, the number of similar cases involved, the number of similar cases handled in different courts, and the number of similar cases that were successful or unsuccessful and their judgments. Fa Xiaotao uses AI to identify the case, and uses big data to retrieve and feedback the above information. 

‘Anhu Tong Al Fabao’ (案沪通AI法宝) is a generative AI service tool launched by the Shanghai High People’s Court and deployed within the Shanghai region, primarily designed to provide litigants with immediate legal information and general legal guidance. The tool is intended to enhance public access to legal information through intelligent means; however, its functional scope is expressly limited to general consultations and informational assistance.

Victims 

A victim is an independent primary participant to criminal proceedings under China’s Criminal Procedure Law. They have direct standing and enjoy similar procedural rights to participate in the criminal process as defendants. They are entitled to retain lawyers to represent them in criminal proceedings.

Administrative support

The 12368 Intelligent Voice System discussed above, is also accessible to victims. By calling the 12368 hotline, victims can make enquiries about case progress and other procedural matters, with information provided either through an AI response or, where necessary, via transfer to a human operator.

Legal research, analysis and drafting support

The National Judicial AI Platform, discussed above, will provide public legal services after the platform is further trained and optimised. It will, for example, answer non-professional queries on legal issues, allowing individuals to access legal services and consultations more easily.

‘AI Fabao (Anhu Tong)’, discussed above, can also provide victims with immediate legal information and general legal guidance.

TRAINING

Although there is no unified national training session for criminal justice actors at this stage, judges may receive training organised by their respective local courts, prosecutors by their local procuratorates, and defence counsel by their law firms or local lawyers’ associations. Such training varies by region and is generally optional. Based on publicly reported training sessions (see also here), the content often includes familiarisation with AI-assisted tools in judicial practice, awareness of their limitations, and emphasis on the principle that AI may support but not replace decision-making. 

As at March 2026, certain sectors have also been progressively incorporating training on the identification of and response to deepfakes into their practical programmes, spanning courts, procuratorates, judicial appraisal bodies, and professional training institutions.

REGULATION 

As at March 2026, there is no nationwide regulation expressly governing the use of AI in criminal or court proceedings in China, but the Supreme People’s Court issued guidance on the development, deployment, and use of AI within the judiciary. In addition, AI-related regulations and policies that are not specific to criminal proceedings, as well as existing laws – such as criminal procedure rules and data protection legislation – apply or offer guidance on the use of AI in criminal proceedings or courts, more broadly. 

AI regulation

While there is no single, comprehensive AI regulation, China has issued a series of AI-related policies and regulations which may provide guidance for AI use in criminal proceedings. 

In January 2022, China introduced the Provisions on the Management of Algorithmic Recommendations in Internet Information Services (2022), which apply to companies offering algorithm-driven services, including content recommendation, ranking, and personalised feeds.

In addition, the Provisions on the Administrations of Deep Synthesis of Internet-based Information Services (2023), which regulates so-called ‘deep synthesis’ technologies such as deepfakes and voice synthesis. It requires providers of deep synthesis services to attach identifiers to content generated or edited using their services, and – where the deep synthesis services may cause public confusion or misidentification – to apply a conspicuous label.

Another recent significant measure is the Interim Measures for the Management of Generative Artificial Intelligence Services (2023), which regulates ‘services to the public in the [mainland] PRC for the generation of text, images, audio, video, or other content’, requiring providers of such services to conduct security assessments, ensure data quality, respect intellectual property, and prevent discrimination and misinformation. Article 12 further provides that ‘Providers shall label generated content such as images and video in accordance with the Provisions on the Administration of Deep Synthesis Internet-based Information Services.’

The Measures for Labeling AI-Generated or Synthesized Content (2025) jointly issued by the Cyberspace Administration of China and three other regulators in March 2025 and effective 1 September 2025, further refine these labelling obligations and extend them to a broader set of regulated actors. The 2025 Measures impose obligations on four categories of entities: (i) AI content generation service providers, (ii) internet-based information content propagation service providers, (iii) app distribution platforms, and (iv) users.

Guidelines for practitioners

Judicial Guidelines 

In 2022, the Supreme People’s Court – the highest judicial organ in China – released the Opinions on Regulating and Strengthening the Applications of AI in Judicial Fields, encouraging the ‘in-depth integration’ of AI to improve the functioning of the judicial system, including with regard to ‘adjudication and enforcement, litigation service, court management, as well as social governance facilitation’; to ‘deepen the construction of smart courts’ and to ‘achieve a higher level of digital justice’. In this publication, the Supreme People’s Court provides guidance on how AI may be developed, deployed, and supervised within the judiciary.

The guidance encourages the following use cases: 

The guidance sets out the following key principles that should govern the deployment and use of AI:

Human oversight and accountability: The guidance ‘affirm[s] the supportive role of AI in adjudication, and the user’s rights to decision-making’. It further states that ‘AI shall not make judicial decisions substituting for the judge in any case, disregarding technological advancement’ and that ‘all judicial powers [must be] administered by adjudicative authorities, and all judicial accountability ultimately falls on the decision-maker’. Users have the right to opt out of interactions with judicial AI services. 

Fairness and justice: The guidance provides that AI must uphold the fairness of the court process and avoid discrimination and prejudice. Courts must follow ‘fundamental judicial rules’, ‘serve […] judicial fairness’, and provide adequate ‘necessary assistance to communities in difficulties and people with special needs to participate in judicial activities [ensuring] universal inclusion of all groups of users with equal opportunities’. 

Transparency and credibility: Courts must ensure ‘the transparency of technology development, product applications, and service operation’. Data ‘collection and management patterns’, ‘the process of legal cognitive semantics processing’, and ‘the logic of assisting judicial presumptions’ must be open ‘to examination, evaluation and registration with the relevant authoritative entities’ and be interpretable, testable, and verifiable. The capabilities as well as limitations of ‘judicial AI products’ must be ‘instructed and identified in a manner that can be easily understood . . . to ensure that the procedure and outcome of applications are predictable, traceable, and credible’. The guidance does not specify whether the use of AI must be disclosed to the parties concerned (e.g., litigants). 

Security and legality: The use of illegal AI technologies and products is prohibited. AI systems must be developed and operated legally, protecting state secrets, network security, data security and data privacy. 

Abiding by public order and good customs: ‘Core Socialist Values’ must be infused ‘into the whole process of technology development, product application and operation of judicial AI’ and a risk management system must be established to ‘resolve possible moral and ethical risks’. 

Moreover, courts are required to put in place top-level design and standards, data infrastructure, research and development priorities, security operations, and oversight and compliance mechanisms to ensure safe and effective deployment of judicial AI. ‘[T]hrough mechanisms such as Judicial AI Ethics Council, People’s courts shall comprehensively adopt methods, including ethical reviews, compliance reviews, and security assessments, to prevent and mitigate cybersecurity risks in judicial AI applications.’

The Supreme People’s Court’s guidance is not an enforceable regulation, functioning primarily as a policy guideline for courts nationwide. Although the guidance does not address the question of non-compliance, judges are subject to the Judges Law, which prescribes sanctions for improper conduct. In the absence of specific rules, these general provisions would apply to, for example, the misuse of AI. Within the judicial administration, there are internal disciplinary proceedings. Every court establishes judge evaluation committees to assess its own judges. These assessments examine judicial performance, professional ethics, expertise, work capabilities, and judicial conduct. Additionally, the Supreme People’s Court and courts at the provincial level establish disciplinary committees that determine whether judges have ‘deliberately violated laws and regulations when handling cases’ or ‘caused serious consequences through gross negligence leading to erroneous judgments’.

The 2022 guidance followed earlier opinions by the Supreme People’s Court, promoting the construction and accelerated implementation of smart courts.  In July 2025, the Supreme People’s Court delivered a high-level lecture on AI integration in judicial proceedings. According to publicly available reports, the Supreme People’s Court highlighted ‘AI’s role in advancing [the] rule of law in the digital era’ and modernising China’s legal system as well as the ‘importance of judicial safeguards in AI application ...  and the challenges that come with AI-powered administration of justice’.

In September 2025, the Xiamen Maritime Court launched Guidelines for Litigation Participants’ Use of Artificial Intelligence (For Trial), following its landmark ruling on counsel’s use of AI in preparing briefs and exhibits in civil litigation. While the original text of the guidelines does not appear to have been published yet, the guidelines reportedly provide that: (1) while litigants are generally permitted to use AI, its use to generate, fabricate, alter, tamper with evidence, or distort factual circumstances is strictly prohibited; (2) litigants must validate all AI-assisted documents and assume full responsibility for the authenticity and accuracy of materials submitted to the court; and (3) when litigants employ AI tools in connection with evidence, material facts, or matters potentially affecting adjudicatory outcomes, they must proactively disclose the precise AI-generated content portions, along with comprehensive details, including AI tool specifications, usage parameters, input data and prompts, and the verification methodologies and procedures.

Bar Association Guidelines

The Guidelines on the Use of AI Tools by Lawyers Providing Community (Village) Legal Advisory Services (2025) (Trial) (‘Guidelines’) were issued by the Shanghai Bar Association for a one-year trial period. They are not mandatory rules, but a practice-oriented reference document, intended to guide Shanghai lawyers in the prudent, compliant, and responsible use of AI tools in community (village) legal advisory services amid the growing adoption of AI in public legal service settings.

Structurally, the ‘Guidelines’ follow a ‘risk identification – principles – operational norms – scenario-based guidance’ framework. They identify key risks associated with lawyers’ use of AI in activities such as legal outreach, legal consultation, and document drafting, including data and information security, the accuracy of AI-generated outputs, algorithmic bias, intellectual property ownership, and potential erosion of professional competence. In response, the ‘Guidelines’ set out core principles – lawfulness and compliance, protection of clients’ interests, privacy and data security, and professional responsibility – and offer relatively concrete, operational recommendations on issues such as AI tool selection, information handling, quality control, record-keeping and archiving, and the labelling of AI-generated or synthesised content, taking into account the characteristics of generative AI and deep synthesis technologies.

While the ‘Guidelines’ recommend obtaining clients’ prior consent before submitting client information to cloud-based AI tools, they do not establish binding disclosure obligations, nor do they purport to create mandatory professional standards or customs.

Overall, the ‘Guidelines’ reflect an anticipatory, self-regulatory response to the risks and boundaries of AI use in legal practice, emphasising the auxiliary role of AI tools and the lawyer’s ultimate professional responsibility, and provide a useful reference for understanding prevailing professional consensus and risk-control approaches to AI use in Shanghai’s public legal services.

Similarly, on 27 October 2024, the Beijing Lawyers Association issued the Research Report on the Impact of Artificial Intelligence on Lawyers’ Legal Services (the ‘AI Research Report’).

In the section titled ‘Recommendations on the Standardised Use of Artificial Intelligence,’ the ‘AI Research Report’recommends that lawyers using AI tools should make adequate disclosure to clients and obtain informed consent. This includes explaining the purpose and intended use scenarios of the AI tools, associated costs, potential risks, and any possible impact on the matter. It further suggests that, upon completion of the engagement, lawyers should clearly indicate whether and how AI tools were in fact used in providing the services. For decisions that may materially affect the client, the AI Research additionally advises that, following full disclosure, the client should determine whether to adopt advice generated by AI tools.

It should be emphasised that these recommendations are framed as practice-oriented guidance and do not have binding force. Accordingly, the ‘AI Research Report’ does not, in itself, establish any mandatory rule or industry practice requiring lawyers, when providing legal services, to disclose their use of AI tools to clients.

Criminal procedure rules 

Admissibility of evidence 

The Criminal Procedure Law establishes standards for evidence admissibility (e.g., requiring that evidence is authentic, relevant, and legally obtained) which would also apply to AI-generated or AI-analysed evidence.

Similarly, the Rules of Criminal Procedure for the People’s Procuratorates (requiring that investigations and prosecutions comply with procedural legality and evidentiary rules); the Provisions on the Procedures for the Handling of Criminal Cases by Public Security Authorities (setting standards for law enforcement’s evidence collection and handling), the Provisions on Several Issues Concerning the Collection, Extraction, Review, and Judgment of Electronic Data in Handling Criminal Cases (setting requirements for the integrity and authenticity of electronic data evidence), and the All-China Lawyers Association’s Rules for the Handling of Criminal Cases by Lawyers (including obligations of confidentiality, diligence, and independence) would also extend to the use of AI.

Generative AI has increasingly been used in criminal activities such as fraud, fabrication, and intellectual property infringement, exposing gaps in existing evidentiary frameworks. In response, the procuratorate and public security bureau of Shanghai’s Xuhui District in 2025 published the Guidelines on the Collection and Review of Electronic Data in Criminal Cases Involving Generative Artificial Intelligence in an effort to address emerging challenges in the collection and review of evidence in generative AI-related crimes. These challenges include: (1) the opacity of generative AI systems, which disrupts causal chains and makes it difficult to establish the link between a specific unlawful act and a particular technical process; (2) the technical complexity of AI-related electronic data, which spans the full chain of model development, service provision, and end use, creating a significant cognitive gap for investigators tasked with identifying, preserving, and interpreting evidence; and (3) the diversity of evidence forms generated by AI systems, including code, logs, model outputs, and user interaction records, which resist straightforward classification and present difficulties in determining applicable review standards.

The Guidelines are reportedly organised into four chapters comprising 42 articles, forming a comprehensive framework that spans foundational norms, evidence collection guidance, review standards, and inter-agency cooperation mechanisms. Chapter 1 sets out the general principles. Chapter 2 addresses evidence collection across four categories of generative AI-related criminal cases, identifying the relevant obligated entities — technology developers, service providers, and technology users — and specifying key collection methods, investigative experiments, and technical appraisal approaches for each. Chapter 3 establishes the standards and methods for procuratorial review of electronic data, emphasising comprehensive review of the multiple connections between electronic data and case facts, and cautioning against isolated assessments. Chapter 4 addresses inter-agency cooperation, calling for the establishment of long-term collaboration mechanisms between public security organs and procuratorates in combating generative AI-related crimes and managing associated security risks. More specifically, in addressing the problem of the “algorithmic black box” being difficult to trace, the Guidelines innovatively proposes experimental methods such as variable comparison, environment simulation, and identity verification to reconstruct data generation pathways, providing operable evidence collection methods for judicial practice.

Disclosure 

China’s Criminal Procedure Law 2018 safeguards the accused’s right to disclosure and prevents selective disclosure by the prosecution. China’s Criminal Procedure Law and its interpretation require the procuratorate to transfer both incriminating and exculpatory evidence, including evidence relevant to the gravity or mitigation of the offence, and allows defence counsel to apply for the retrieval of any exculpatory or mitigating evidence not transferred. The court is responsible for reviewing the completeness of the evidence and may order supplementary disclosure.

Defence counsel may adduce evidence favourable to the defendant at trial without prior notice to the prosecution. China’s Criminal Procedure Law also provides for pre-trial conferences, encouraging the prosecution and defence, under judicial supervision, to exchange evidence lists and address issues such as the exclusion of illegally obtained evidence, in order to improve efficiency and transparency.

As at March 2026, there are no laws that expressly require prosecutorial authorities or other law-enforcement and case-handling bodies to inform parties (including their legal representatives) that artificial intelligence tools or systems have been used in the handling of a specific case.

Deepfakes 

As at March 2026, Chinese criminal law has not introduced a standalone offence specifically targeting deepfakes. Nevertheless, conduct involving harmful uses of deepfake technology can be accommodated within the existing framework of criminal offences. By way of example, such conduct may fall within offences in the Criminal Code, including fraud (Article 266) and the unlawful infringement of citizens’ personal information (Article 253 (I)).

In response to the risks posed by deepfake evidence, both academic commentators and practitioners have proposed a range of reforms to evidentiary rules and corresponding technical countermeasures. For example:

• Flexible allocation of burden of proof: To address the deadlock of uncertainty as to authenticity caused by deepfake evidence, some scholars propose adjusting the allocation of the burden of proof under specific circumstances (for instance, by requiring the party submitting the evidence to further prove its authenticity). Similarly, some commentators advocate shifting the principles of evidence review: certain types of electronic evidence that are easily falsifiable (such as audio recordings) may be presumed inauthentic, with the submitting party bearing the burden of proving their validity and subject to stricter admissibility standards.

• Introduce expert assistance: Some proposals advocate enhancing the expert assistance system by appointing digital forensics and algorithm specialists to aid courts in complex cases, accompanied by clearer standards for admitting expert testimony.

• Enhance technical countermeasures: Some criminal justice practitioners have also suggested that reviewers should scrutinise deepfake evidence for inherent technical imperfections, such as unnatural movements, facial inconsistencies, and distorted hands, and examine encoding parameters and bitrate for anomalies. On provenance and preservation, original storage media should be seized where possible, and metadata such as file format, resolution, and timestamps should be cross-verified against source data to detect signs of modification.

Perceived need for AI-related reform in criminal procedure

More generally, current discussion and practice increasingly reflect a clear recognition of the need for reform, primarily in the following respects:

• Development of rules and guidance for AI-assisted evidence review. There is a recognised need to establish evidentiary review rules aligned with the use of AI technologies, clarifying conditions of use, procedural requirements, and standards of responsibility to ensure legality and procedural integrity. The existing framework contains significant gaps. For example, AI evidence is not formally recognised as a distinct legal evidence category under PRC law, creating uncertainty as to applicable classification, admissibility standards, and review criteria. In addition, various factors, including errors, algorithmic bias, and algorithmic black boxes, affect the reliability of AI evidence, yet there is no dedicated framework addressing these concerns.
• Supplementation of electronic data collection and review rules in generative AI-related cases. As discussed above, Guidelines on the Collection and Review of Electronic Data in Criminal Cases Involving Generative Artificial Intelligence jointly issued by the procuratorate and public security bureau of Shanghai’s Xuhui District are widely regarded as constructive efforts to address emerging challenges in the collection and review of evidence in generative AI-related crimes. These localised practices, in turn, point to the need for a more systematic, national-level regulatory framework.
• Limitations of existing norms and the case for legislative response. The Opinions on Regulating and Strengthening the Applications of AI in Judicial Fields is a policy document rather than binding law and its impact is therefore limited in terms of legal authority, enforceability, and practical reach. Consequently, some commentators advocate incorporating AI-related provisions into future amendments to the Criminal Procedure Law, in order to provide a more stable and comprehensive response to challenges arising in digitalised criminal proceedings.

Data protection legislation 

The Personal Information Protection Law (2021) requires that personal information be processed on a lawful, justified, and necessary basis, with safeguards such as data minimisation and security protections, allowing certain exceptions for state organs performing statutory functions, including in criminal procedure. These would also apply to AI tools used in criminal cases to process sensitive data.

Similarly, the Data Security Law (2021) mandates classification, protection, and security assessment of data, especially ‘important data’. Criminal case files and evidence processed by AI systems may fall within this scope, meaning such systems must operate under strict security and governance obligations. These requirements apply both to parties in criminal proceedings handling judicial data and to third-party technology providers involved in processing such data through AI applications. 

The Artificial Intelligence Security Governance Framework may also provide high-level guidance for AI use in criminal proceedings. Appendix 2 of this guideline, for example, requires that a human control system should be established at critical stages of AI systems to ensure that humans retain the final decision-making authority. Measures include designing safety control thresholds, and reserving an effective window for human intervention, so that AI systems can achieve intended human objectives and do not operate uncontrollably without human oversight.

Human rights 

Guarantees, including the right to a fair trial and privacy rights, under other international human rights treaties to which China is a party, such as articles 16 and 40 of the Convention on the Rights of the Child, may also be relevant to the use of AI tools in court.

Outlook 

The Standing Committee of the National People’s Congress’s 2024 legislative plan included the ‘healthy development of artificial intelligence’. The Global Artificial Intelligence Governance Initiative, which was released in October 2023, includes suggestions to ‘gradually establish and improve relevant laws, regulations and rules, and ensure personal privacy and data security in the R&D and application of AI’.

CASES

As at March 2026, a small number of civil cases have begun to address the use of AI-generated or AI-manipulated materials in litigation, including in relation to evidence and legal submissions.

Misuse of AI in court filings

In 2025, reportedly for the first time in Chinese judicial practice, a regional court addressed counsel’s use of AI in preparing briefs and exhibits in civil litigation, requiring them to fully disclose the use. This landmark ruling may also influence AI applications in criminal proceedings.

The Xiamen Maritime Court had to determine the validity of an arbitration agreement and noted that claimant’s counsel may have used AI when preparing the court submission. Drawing on the Chartered Institute of Arbitrators’ Guideline on the Use of AI in Arbitration (2025), the Court emphasised the importance of disclosing AI usage to ensure fairness, transparency and the integrity of the proceedings, and held that when AI intervention may materially impact case substance – including evidence and arguments – users must provide full disclosure to the court. The Court required counsel to disclose:

1. the specific AI tools used;
2. the scope of the AI use (such as evidence summarisation, legal research, document drafting) as well as the intended purpose;
3. the documentation of the use (with precise identification of the portions of evidence, legal arguments, or other litigation documents for which AI was used for);
4. assurances that all AI-assisted materials have undergone final review and verification by the lawyer or appropriately qualified personnel; and
5. a commitment to strict adherence to all applicable laws and regulations governing data security, personal information protection, and confidentiality obligations throughout AI use.

This ruling of the Xiamen Maritime Court directly catalysed this court’s launch of the above-mentioned Guidelines for Litigation Participants’ Use of Artificial Intelligence (For Trial) in September 2025.

Finally, courts have also addressed the misuse of AI in legal argumentation. In one case, plaintiff’s counsel cited AI-generated reference cases in written submissions that appeared complete but did not exist in any recognised judicial system and had not been verified. The court found that this conduct amounted to fabricating judicial authorities and misleading the adjudicative process, and expressly criticised it in its judgment.

Deepfakes and synthetic media

Beyond this institutional response, several civil cases illustrate more direct attempts to use AI-generated or AI-manipulated materials as evidentiary support. In a residential lease dispute (The Primary People’s Court of Dawu County of Xiaogan City, Hubei Province; December 2025), the claimant submitted photographs of utility meters to prove unpaid charges; however, the images bore a visible ‘AI-generated’ watermark and were later admitted to have been created using AI tools. The court held that the photographs constituted fabricated evidence, refused to admit them, and formally admonished the claimant.

In a separate lease dispute (The Primary People’s Court of Luqiao District of Taizhou City, Zhejiang Province; December 2025), a party submitted screenshots of WeChat chat records to demonstrate the opposing party’s agreement to pay certain fees, but discrepancies were identified when the court examined the original messages. After the party admitted to using AI to alter the chat content, the court treated the conduct as evidence fabrication and imposed a monetary fine.