
SHI Jingxia
Wu Yuzhang Chair Professor of Law School, Renmin University of China
Distinguished colleagues,
It is my great honor to participate in this symposium of the International Commercial Expert Committee of the Supreme People’s Court. Under Topic Four, I would like to discuss an issue that may appear technical, but is in fact already beginning to affect the administration of international commercial justice: the risks posed by artificial intelligence “hallucinations” in legal proceedings and how we should respond to them.
Let me begin with a recent real-life case. In April 2026, in a Chapter 15 filing before the U.S. Bankruptcy Court for the Southern District of New York concerning an application for cross-border insolvency recognition and assistance, the counsel from a prominent U.S. law firm cited a large number of judicial decisions in a motion filed with the court. It was subsequently discovered that a significant number of those citations were erroneous and had been generated by artificial intelligence. The law firm later apologized to the court, acknowledging that at least dozens of errors had originated from AI-generated content and that its internal verification mechanisms had failed to function effectively. This case sends a clear signal: AI is no longer something that “may affect” the judicial process; it has already entered the judicial process. The impact of AI is particularly significant in the field of international commercial adjudication for three reasons. First, international commercial cases are inherently multi-jurisdictional. When AI is asked to deal simultaneously with different legal systems, languages and sets of rules, it can easily piece together an answer that appears entirely plausible but is in fact incorrect. Second, international commercial legal documents rely heavily on case citations. This is particularly true in common law jurisdictions, where case citations are extensive. Yet this is precisely where AI hallucinations are most likely to occur: AI may simply “invent” a case that does not exist, while presenting it in a format that looks perfectly authentic. Third, and most importantly, courts and arbitral tribunals rely to a considerable extent on materials submitted by the parties. This means that if AI-generated errors find their way into party submissions and remain undetected, they may ultimately enter the adjudicative process itself.
My central point, therefore, is that AI hallucination is not fundamentally a technological problem. It is a problem of evidence and procedural trust. It affects the authenticity of evidence, the reliability of legal reasoning, and the fairness of proceedings. In other words, it touches upon the very foundations of the judicial process.
Can our existing legal and procedural frameworks adequately address the improper impact of AI on judicial proceedings? My view is that, at present, they are probably not sufficient. Lawyers are, of course, already subject to duties of diligence and good faith. But these duties were developed primarily to address human error rather than machine-generated error. What distinguishes AI-related risks in international commercial justice is that errors may be more difficult to detect, may occur on a much larger scale, and, may be more readily trusted. Traditional professional obligations may therefore prove insufficient in practice to address these new risks.
Against this background, I would like to offer three suggestions for discussion.
First, we should strengthen the duty to verify materials submitted in international commercial legal proceedings. It may be appropriate to make clear, as a matter of practice, that AI-generated content must be independently verified. In certain circumstances, the submission of unverified AI-generated material might even be treated as a procedural deficiency. Second, we could consider introducing a proportionate disclosure mechanism. In appropriate circumstances, parties or counsel could be required to disclose whether AI has been used and, if so, the extent of that use. The purpose would not be to restrict the use of AI in international commercial legal proceedings. Rather, it would be to enhance transparency and assist courts in assessing the reliability of the materials before them. Third, we should explore the development of guidelines for the use of AI in international commercial justice. Such guidelines could address, for example, standards for reviewing AI-generated materials, requirements concerning the cited sources, and mechanisms for identifying anomalous citations. These standards could gradually be developed through judicial guidance or rules of practice.
Let me conclude with one brief observation. There is no doubt that artificial intelligence will improve the efficiency of international commercial justice, and this is an irreversible trend. At the same time, without appropriate rules and safeguards, AI-generated errors may be amplified and find their way into the judicial system. The challenge before us, therefore, is not whether AI should be used in international commercial justice. The real question is how we can make use of AI while safeguarding the accuracy of judicial decision-making, the integrity of due process, and public confidence in the administration of justice.