Most South Korean judges encounter AI-generated fake legal precedents
A Supreme Court survey of 360 judges found 80.7 percent had seen court cases citing nonexistent legal precedents, and nearly 92 percent want institutional measures.
Eight in 10 South Korean judges have encountered court cases citing AI-generated fake legal precedents, a Supreme Court survey released on Monday showed, South Korea's national news agency Yonhap reported.
The court surveyed 360 judges from Oct. 6 to 16 to examine what it called "AI hallucination," a term for cases in which AI models present false information as if it were true.
Of the judges handling civil, family and administrative cases, 196, or 80.7 percent, said they had seen citations of nonexistent legal precedents, while 159, or 65.4 percent, said they had encountered citations of false precedents.
Smaller shares reported other kinds of fabricated material, with 41 judges, or 16.9 percent, citing false or nonexistent laws and regulations, and 10 judges, or 4.1 percent, citing false theses or publications.
The survey also found that 164 judges had themselves cited false precedents, while 136 judges, or 57.6 percent, said lawyers representing parties in court had cited false precedents or other information.
Nearly 92 percent of respondents said institutional measures are needed to address citations of AI-generated fake precedents.
Those measures include setting up legal grounds for imposing fines, establishing verification requirements and devising disciplinary measures against lawyers who repeatedly cite false information.
An official at the National Court Administration said citing AI-generated false or nonexistent precedents may delay court proceedings, and that the survey results should guide effective countermeasures.
Generative AI tools can produce text that reads convincingly but contains inaccurate or invented sources, a problem that has drawn concern among legal professionals in several countries.
