No. 107
2026-11-03
the AI pulse — brewed, not raw
 AI CRITIQUE
Görselde bir çekiç, üzerinde kırmızı bir 'X' bulunan bir belge, para yığınları ve Güney Kore'nin Seul şehrinde binalar ve dağlardan oluşan bir arka plan yer almaktadır.

South Korea seeks fines for AI-generated fake case law

Read the article    English (automatic translation)

Summary

South Korea's National Court Administration reported that approximately 80% of judges in civil, family, and administrative courts have encountered petitions citing court decisions that were fabricated or distorted by artificial intelligence. The agency aims to establish a legal basis for imposing administrative fines (과태료) through amendments to both civil and criminal procedural laws.

Why it matters

This marks one of the first major judicial systems to officially measure the scale of AI hallucinations in court filings and initiate a legal response. This step creates a potential model for other jurisdictions grappling with the same issue. The survey revealed that 57.6% of fake citations originated from lawyers, not just pro se litigants, shifting the regulatory debate toward professional accountability. However, since the fine amounts, timeline, and enforcement authority have not yet been determined, this move resembles a statement of intent more than a functional deterrent.

Highlights

  • 360 judges were surveyed. 80.7% reported encountering citations of non-existent decisions, while 65.4% encountered citations of distorted decisions. Some petitions even contained raw chatbot conversation snippets pasted into the text.
  • 69.5% of judges indicated that fake references were brought by parties to the case, while 57.6% indicated they came from lawyers.
  • 91.7% of participants called for an institutional response. However, the administration's announced follow-up plan includes an AI disclosure obligation, which was supported by 60.6% of judges.

Critical take

The survey methodology is weak. The number of judges who completed the survey and the response rate were not disclosed. It is also unknown which pool the 360 responses came from. Therefore, the 80% headline figure should be read with caution. Additionally, the administration's own follow-up plan quietly downplays the mandatory disclosure of AI use and reduces the verification obligation to merely a reminder. This suggests that the legal initiative is currently based more on show than substance.

This critique was written by the local AI model Qwen3.8-27B. I chose the story; I do not review each critique before it is published.
Back to the issue