◆ Deep forgery relies on artificial intelligence algorithms, and its technical traces are more concealed than traditional editing. It significantly increases the technical threshold and proof cost of evidence authenticity review, and can easily lead to deviations in judicial personnel's evidence review and fact determination.
◆There is a trend that deserves alertness in current litigation practice: parties in a disadvantaged position in litigation file a defense against real electronic evidence on the grounds that the video or recording is forged by AI. This has significantly increased litigation costs and significantly reduced judicial efficiency.
◆Faced with the impact of new technologies, judicial staff should no longer "treat headaches and foot problems". Experts suggest that a stepped classification review mechanism and a full-chain evidence review model should be established. Refining the evidence review rules and accountability system is the last line of defense against the influx of false evidence into the judicial process.
At the end of 2025, the owner of a restaurant in Yangpu District, Shanghai received a diagnosis certificate and vomit photo from a customer, Yang, saying that he suffered from acute gastroenteritis due to eating in the store and claimed 2,000 yuan in "medical expenses."
The store lost money, but it wasn't until the police came to find out that the diagnosis certificate and realistic photo of vomitus were forged by Yang. Yang entered a few lines of text into the AI software, and in just a few seconds, the "diagnosis" was generated.
Although the case is small, it raises a new proposition of the era: with the popularity of generative artificial intelligence, the cost of forging evidence is so low that it can be ignored. From being "operated" by professionals to being able to be generated by everyone with one click, the threshold for counterfeiting has been greatly lowered.
When bank receipts can be fabricated out of thin air using codes, when the faces of people in surveillance videos are reassembled by algorithms, when the reference cases submitted by lawyers are just an "illusion output" of a large language model, the real evidence on which judicial operations depend is being quietly eroded in the AI era.
Under such circumstances, how can the judicial organs maintain the bottom line of fairness and justice?
Deep fakes mislead evidence judgment
At the end of 2025, Yang used AI software to generate acute gastroenteritis diagnosis certificates and realistic vomit photos after eating in or buying takeout at multiple restaurants. Subsequently, he threatened to complain to the market supervision department and write a negative review on the consumer platform, and demanded "medical expenses" from many restaurants.
In June this year, the Yangpu District Procuratorate prosecuted Yang for using AI technology to forge medical materials, make malicious complaints, and extort merchants. After a trial, the court sentenced Yang to four months of detention for extortion, suspended for four months, and fined him 2,000 yuan.
"If it were not carefully compared with the diagnosis certificate issued by the hospital, it would be difficult to find that it was generated by AI." Qian Wen, prosecutor of the Yangpu District Procuratorate, told reporters.
"In the past few years, people were still ridiculing AI for being 'mentally retarded'. Now, it has become an out-and-out AI 'painter', and the content it generates is sophisticated enough to be fake." Qian Wen said frankly that this brings a lot of challenges to judicial staff in reviewing evidence.
Challenges are not limited to the criminal field. In a house rental dispute case, the lessee Han submitted several electronic bank quick payment receipts with standardized formats and clear seals to the court of first instance to prove that the rent had been paid, and the court of first instance accepted this. However, after the other party appealed, the situation took a dramatic turn. The judge of the second instance court obtained bank account transaction records and discovered that the transfer records provided by Han were actually forged using AI technology. Later, the court of second instance changed the judgment of the case.
"In the past, when identifying the authenticity of audio-visual materials, we mainly focused on whether there were traces of traditional processing such as editing and splicing. Such methods usually leave identifiable technical features." A technician from the Procuratorial Technology Information Research Center of the Supreme People's Procuratorate pointed out that deep forgery relies on artificial intelligence algorithms, and the technical traces are more concealed than traditional editing, which significantly increases the technical threshold and proof cost of evidence authenticity review.
The above-mentioned technicians also pointed out that parties can use AI technology to "tailor-made" forged evidence according to the needs of litigation, and even construct a complete chain of false evidence. Because forged evidence may form seemingly "reasonable" mutual corroboration with other evidence, it can easily lead to biases in judicial personnel's evidence review and fact determination.
Zheng Fei, a professor at the Institute of Evidence Science of China University of Political Science and Law, pointed out that once deeply forged evidence materials are used as the basis for finalizing a case, it will directly affect the judge’s judgment of the facts of the case, leading to wrong cases and damaging judicial credibility.
Traditional evidence review methods face unprecedented challenges
What's more troublesome than "difficult to prove and falsify" is that this crisis of trust is spawning a reverse litigation speculation.
Zheng Fei revealed to reporters that there is a trend worthy of vigilance in current litigation practice: parties in a disadvantaged position in litigation argue against real electronic evidence on the grounds that the video or recording was forged by AI. As the saying goes, "If you lie with your mouth, you will lose your legs if you refute rumors." This has significantly increased litigation costs and greatly reduced judicial efficiency.
The trouble caused by deep forgery sometimes does not come from the malicious intent of the parties, but may also stem from the unintentional mistakes of legal practitioners.
In 2025, the Tongzhou District Court of Beijing heard a civil dispute arising from agency shareholding. The plaintiff's attorney submitted a written representation after the court, accompanied by two reference cases that "perfectly supported" his point of view. When reviewing the case, Judge Zheng Jizhe vaguely felt that the content of the case was too consistent with the facts of the case. After verification, he found that the content of the two cases was completely inconsistent with the actual cases corresponding to the corresponding case numbers. Upon inquiry, the plaintiff's attorney admitted that these two cases were obtained after asking questions to an AI, and were written into the attorney's opinions without verification.
Since the punishable forgery under the Civil Procedure Law mainly targets the evidence itself, and the reference cases in the attorney's opinion do not belong to the category of statutory evidence, the attorney only received criticism and education. This incident undoubtedly sounded the alarm for legal practitioners - where is the boundary of the assistive role of AI tools? Where is the bottom line of professional responsibility?
The reporter learned from interviews that most lawyers believe that AI tools can effectively improve work efficiency in daily work, and the benefits outweigh the disadvantages. When asked how to treat AI-generated content, Liu Yanyan, a lawyer at Weiheng Law Firm, said frankly that she would treat AI-generated content with caution. "For example, in case retrieval, AI will be asked to provide specific case numbers and then be manually verified one by one. This should be a lawyer's professional consciousness."
In judicial practice, images, audio and video have always been regarded as objective evidence with strong authenticity and high reliability. However, when an algorithm can generate any video or audio recording out of thin air, the authenticity of the evidence becomes uncertain from the source.
“This declares that the traditional ‘observation-empirical review method’ is basically ineffective,” said Pan Jingui, a professor at the School of Law at Southwest University of Political Science and Law.
How to deal with AI deepfakes
Faced with the impact of new technologies, judicial staff should no longer "treat the head when it hurts and the foot when it hurts".
Zheng Fei suggested that a stepped classification review mechanism should be established: the first step is to "look at the level". The higher the level of evidence generated by AI, the more low-level evidence such as original electronic data must be used for cross-verification; the second step is to "judge the difficulty", which requires routine review of simple evidence, while complex ones require the help of technology. Technical means are used to assist judgment. For those that cannot distinguish true from false, we would rather exclude the doubtful evidence than allow it to enter the finalization process; the third step is "substantial review", which combines the testability of large model algorithms and the rationality of generated logic to conduct a deeper review of the relevance, authenticity and legality of the evidence.
For example, in 2024, Wang Moumou used audio and video materials downloaded from the Internet to fabricate a drunken speech script by Lu Moumou, the founder of Hefei Sanyang Network Technology Co., Ltd. (hereinafter referred to as "Three Sheep Company"), and used AI tool training to generate fake audio. After receiving the report from Three Sheep Company, the public security organs immediately sealed the original electronic data carrier and found evidence of forged audio and video in Wang's computer, mobile phone and website that produced AI audio, which played a key role in the subsequent handling and characterization of the case.
Pan Jingui suggested from the operational process level to build a full-chain evidence review model of "formal screening-substantive verification-comprehensive verification-expert assistance". He particularly emphasized that evidence must be cross-verified through at least two different technical means, and electronic evidence must be placed in the entire case evidence system to examine whether it can be mutually corroborated with verbal evidence, physical evidence, documentary evidence, and even third-party objective data.
Liu Yanyan also believes that expert assistance plays an important role. She suggested that judicial organs learn from the practice of equipping technical investigators in intellectual property cases and hire technical personnel with professional backgrounds such as artificial intelligence and computer technology to provide professional support throughout the entire process from case filing to court hearing.
However, the introduction of technological means also comes with new worries.
Pan Jingui pointed out that when judicial authorities become "technically dependent" on evidence review, the judgments of relevant technical institutions may essentially replace the judicial authorities' legal judgments, resulting in the judicial power of evidence review judgment and fact determination actually being partially transferred to technical service providers who have the advantages of algorithms and data.
Previously, the Beijing Internet Court heard a short video copyright case. In this case, three authoritative institutions appraised the same artificial intelligence-generated video and came to the appraisal opinions of "artificial creation", "algorithm generation" and "undeterminable" respectively. It can be seen from this that the issues of credibility and standard unification of the technical appraisal system itself cannot be avoided.
Under such circumstances, if one "blindly believes" in the technical appraisal opinions and then "blindly accepts" them, the independence and impartiality of judicial review will inevitably be subject to potential risks.
The judiciary is taking active action
Technical problems require technology to solve, but technology alone is far from enough. Refining the evidence review rules and accountability system is the last line of defense against the influx of false evidence into the judicial process.
Liu Yanyan suggested that for the review of AI deep forgery evidence, judicial authorities should issue special evidence review guidelines and strengthen operational guidelines for case investigators.
In fact, in this regard, the formulation of relevant standards is already on the way. In 2025, with the participation of units such as the Identification Center of the Ministry of Public Security, the Technical Information Research Center of the Supreme People's Procuratorate, and the Forensic Identification Science Research Institute, the national standard "Forensic Science, Forged Portraits, and Deep Forgery Inspection" was officially released by the State Administration for Market Regulation and the National Standardization Administration. The standard specifies in detail deep forgery inspection equipment, inspection steps, and the basis for judgment of appraisal opinions, and provides unified technical specifications for the inspection of deep forgery video images.
Local judicial authorities have also conducted active exploration. In 2025, the Xuhui District Procuratorate of Shanghai and the Xuhui Branch of the Shanghai Public Security Bureau jointly issued the "Guidelines for Electronic Data Collection and Review of Generative Artificial Intelligence Criminal Cases", forming a complete system of basic specifications, evidence collection guidelines, review standards and collaboration mechanisms.
Another key to system building is accountability. It is worth noting that on September 7 this year, the Supreme People's Court issued the "Opinions on the Legal Trial of Cases Involving Artificial Intelligence Disputes" (hereinafter referred to as the "Opinions"), which particularly emphasized that the use of artificial intelligence to improperly collect evidence and other behaviors that hinder judicial order shall be regulated in accordance with the law.
The "Opinions" clearly stipulate that if litigation participants or other persons use artificial intelligence to forge evidence and hinder the People's Court from hearing cases, they shall be handled in accordance with Article 114 of the Civil Procedure Law. If litigation documents, case retrieval reports and other materials submitted by litigation participants are generated using artificial intelligence, they should carefully verify the authenticity and accuracy of relevant laws, judicial interpretations, cases and other content before submitting them to the court, explain the use of artificial intelligence assistance when submitting to the court, and bear responsibility for the authenticity and accuracy of the relevant content in accordance with the law.
The impact of AI on the judiciary is essentially a matter of "time lag" between technological progress and institutional evolution. From national standards to case handling guidelines to the issuance of the Opinions, in the game of responses and challenges, judicial authorities are taking active actions to seek to upgrade and iterate their evidence review capabilities in the AI era.
