I. BACKGROUND AND PURPOSE OF ANNOUNCEMENT

The Court Artificial Intelligence Research Society (the “AI Research Society”) announced the “Guidelines on the Use of Artificial Intelligence in the Judiciary” (“the “Guidelines”) in February 2025. The AI Research Society was established within the court on October 24, 2023, and has been studying systems and issues related to artificial intelligence (“AI”) technology, including how AI can be leveraged to support trial work. The Guidelines outline key principles judges should consider when using AI and other relevant factors regarding AI’s application in trials. 

While the Guidelines do not represent the official stance of the courts, they provide valuable guidance on standards and procedures that may be considered and adopted as AI becomes further integrated into litigation processes.

II. SPECIFIC DETAILS

The Guidelines consist of General Provisions (Chapter 1), Basic Understanding of AI (Chapter 2), Development and Introduction of AI in Judiciary (Chapter 3), Use of AI by Judges (Chapter 4), and Courts’ Response to Use of AI by Litigants (Chapter 5).

1. General Provisions

The Guidelines emphasize that while AI is a useful tool that can enhance the efficiency of judicial processes and improve the people’s access to justice, AI must be employed with caution, in line with the fundamental principles of justice, as AI is fundamentally subject to its inherent limitations. 

2. Basic Understanding of AI

AI is a powerful tool, but it presents significant risks, including (i) the generation of fabricated content (the so-called “hallucination phenomenon”), (ii) bias due to flaws in training data, and (iii) opacity in data generation process and the possibility that AI outputs may be misused, such as through deepfakes.

3. Development and Introduction of AI in Judiciary

The Guidelines establish key principles that the judiciary must adhere to when integrating AI based on such inherent risks in AI. 

• Guaranteeing Fundamental Rights and Constitutional Values: AI in the judiciary should be introduced in a manner that upholds the fundamental rights of the people. Special attention should be given not to undermine the right to be tried by judges (Article 27(1) of the Constitution) and the judicial independence (Article 103 of the Constitution). In addition, to ensure the principle of equality (Article 11(1) of the Constitution), AI’s inherent bias must be mitigated, and due process must be maintained, by ensuring that the litigants’ participation is not unduly restricted. 

• Reliability: AI used in the judiciary must be accurate and equitable, and it needs to be subject to ongoing monitoring and improvement to minimize bias in training data.

• Legality: AI used in the judiciary must align with the overall legal framework of the Republic of Korea, including compliance with the Personal Information Protection Act and intellectual property laws in relation to data collection and output generation. Where necessary, the legal basis and limitations of the judicial AI usage should be codified in the Civil Procedure Rules, Criminal Procedure Rules, and other relevant laws and regulations.

• Responsibility: AI used in the judiciary should be designed to enable judges to verify and assess the accuracy of the outputs.

• Transparency: The judiciary must document AI usage, including data sources, screening processes, and training methods, and make all or part of this information publicly available. 

• Future Orientation: AI should be designed with a forward-looking perspective, reflecting legal judgments and societal changes, and be subject to continuous review and improvement.

4. Use of AI by Judges

A. Guaranteeing Fundamental Rights and Constitutional Values 

The Guidelines emphasize that judges must ensure that their AI usage does not compromise constitutional values and fundamental rights related to the judiciary, including judicial independence, the right to a trial, and due process. Specifically, judges should not rely exclusively on AI to find facts and interpret and apply the law. In particular, judges must provide appropriate and sufficient information regarding their AI usage if AI is used in matters that significantly affects the defendants’ rights to defense and counsel in criminal proceedings. Furthermore, judges should be cognizant of AI-generated biases and scrutinize outputs to ensure the principle of equality is upheld. 

B. Ensuring Responsible Use and Reliability 

Judges must recognize that their reliance on erroneous AI outputs may lead to their legal and ethical liability. Accordingly, judges must possess a basic understanding of AI risks, and always critically evaluate AI-generated content. 

C. Considerations for Commercial AI Use: Protecting Personal Information and Confidentiality

Judges should not input the following information into commercial AI platforms unless both parties explicitly consent: personal information, details of a case, trade secrets, details about the private lives of the people involved in the case, or draft judgments. 
Furthermore, judges should avoid using official judicial email addresses to register for commercial AI services and should not disclose their occupation as judges in their account profiles. If a judge’s personal information is leaked, the judge must promptly contact the judiciary’s technical support team for appropriate technical assistance.

5. Courts’ Response to Use of AI by Litigant

A. Litigants' Obligation to Disclose AI Use

The Guidelines state that courts should direct litigants to clarify their use of AI. Specifically, if the court suspects or the opposing party alleges that a litigant has used AI in producing documents or evidences, the court may demand disclosure of the AI usage. In criminal cases, the admissibility of evidence is stringent, so judges must assess the admissibility of evidence in accordance with the Criminal Procedure Act and the Criminal Procedure Rules. 

The Guidelines recognize that current technologies cannot accurately detect AI-generated outputs, but also point out that AI usage can be suspected if: (i) submitted materials contain obvious errors (e.g., if the cited case number does not exist), (ii) different precedents are referenced for similar legal issues, or (iii) submitted materials deviate from general legal principles. 

Furthermore, if AI-generated documents or evidence are identified, the court may compel the submitting party to disclose: which  AI used, the commands/prompts entered into the AI, and the steps taken by the litigant to verify the AI output, to verify the accuracy. In addition, the Guidelines suggest considering amendments to the Civil Procedure Rules, Criminal Procedure Rules or other relevant rules to mandate the disclosure of AI usage in litigation materials and evidence. 

B. Challenges Posed by Deepfakes

The Guidelines highlight that deepfakes can obstruct the finding of truth. Specifically, the Guidelines note that (i) a litigant may deliberately submit documents and evidence generated by deepfake technology, or (ii) an opposing party may allege forgery or manipulation based on deepfake technology. Given current limitations in detecting deepfakes, courts may face challenges in finding the truth. 

In addition, an appraisal to detect deepfakes can be costly and cause delays in the trial process. Moreover, litigants with limited knowledge of deepfake technology may not be able to respond effectively to contest forged materials, and potentially be compromised of their right to a fair trial. Therefore, the Guidelines urge judges to exercise due regard to the challenges posed by deepfakes in conducting the trials. 

III. IMPLICATIONS

In recent years, the legal profession has been increasingly integrating AI in various ways and scopes. In foreign jurisdictions, some lawyers have even been fined1 or had their licenses suspended2 for submitting AI-generated documents containing false information. 

As AI usage becomes more widespread, disputes over AI-generated materials are expected to rise. The Guidelines provide a valuable framework for navigating these challenges. In future litigation, it will be essential to develop effective methods to identify, address, and challenge AI-generated materials that contain errors or falsehoods. 

Lastly, these Guidelines call for amendments to the procedural rules to account for AI-generated outputs. Given the rapid pace of AI adoption, the principles and rules outlined in the Guidelines may gain momentum in the near future. Therefore, it is crucial to monitor ongoing discussions, evaluate their impact on litigation practices, and develop strategies to address AI-related legal challenges effectively. 

1. Yoon-seop Jung, “U.S. Lawyers Fined for Submitting ‘Fake Precedent’ Written by ChatGPT,” KBS News, June, 23, 2023
2. Soo-jung Hong, “U.S. Court Suspends Lawyer Who Cited ‘AI-generated Fake Precedent’ for 1 Year”, Law Times, March 28, 2024