South Korean judges report high frequency of AI-generated fake legal precedents

A survey conducted by the South Korean Supreme Court revealed that approximately 80 percent of judges have encountered court cases involving false or non-existent legal precedents generated by artificial intelligence. The study, which surveyed 360 judges between October 6 and October 16, aimed to investigate instances of “AI hallucination,” where AI models present false information as fact.

Among the respondents, 80.7 percent reported seeing citations of non-existent precedents, while 65.4 percent encountered false precedents. The survey also noted that 16.9 percent of judges saw citations of false laws or regulations, and 4.1 percent encountered false theses or publications. Notably, 164 judges reported having cited false precedents themselves, while 57.6 percent stated that lawyers representing parties in court had cited false information.

Nearly 92 percent of the judges surveyed indicated a need for institutional measures to address this issue. Proposed solutions include establishing legal grounds for fines, creating verification requirements, and implementing disciplinary measures for lawyers who repeatedly cite false information. Officials from the National Court Administration warned that these inaccuracies could lead to significant delays in court proceedings.