민사·가사 판사 10명 중 8명 “AI 허위 인용 경험”…법원, 과태료 부과 검토

Out of 10 judges handling civil, family, and administrative cases, 8 have reportedly encountered cases with AI-generated, manipulated case precedents. A survey conducted between June 6th and 16th by the Supreme Court's Office of Legal Administration, which polled 360 judges across all courts, revealed that approximately 80% of civil, family, and administrative case judges had experienced such situations.

The most common type of manipulated precedent cited was fabricated existing precedents, accounting for 80.7% of the cases. Next were unoriginal existing precedents at 65.4%, followed by fabricated legal provisions at 16.9%, and finally unoriginal papers or scholarly works at 4.1%. Respondents indicated that 69.5% of false data submissions came from the parties themselves, while 57.6% were submitted by their attorneys or counsel.

The survey also uncovered numerous instances where attorneys submitted fabricated or inconsistent case precedents, including using the Supreme Court's format to generate false rulings or citing non-existent case numbers. Some instances involved directly incorporating the actual case content into the submitted documents, even though the cited precedent could not be found.

Over 91.7% of respondents agreed that "institutional measures are required." The primary proposed measures included imposing fines for using manipulated precedents (61.4%), enforcing verification obligations for cited precedents (61.1%), requiring attorneys to disclose AI usage (60.6%), and disciplinary actions against attorneys (55.4%).

The Legal Administration noted that false citations can unnecessarily burden opposing parties with the need to verify and refute evidence, and can hinder the efficient progress of trials. They indicated that legislative reforms to the Civil and Criminal Procedures Codes, such as establishing guidelines for fines, would be necessary.

They also stated that they would submit comments to the National Assembly in consultation with relevant government agencies.