Legal Affairs

"Nothing but Irrelevant Precedents"… 'Trial Support AI' Ignored by Judges

[Challenges for the Judiciary in the First Year of AI Adoption] ① Supreme Court Administrative Office Accelerates AI Organizational Restructuring and System Development AI System for Court Support Worth 16.1 Billion Won Sees "Plummeting" Adoption on the Front Lines "Budget Shortages Limit GPU and Other Equipment Procurement... Reliance on External Labor" "Need to Enhance Practical Skills, Such as Writing Summaries and Outlining Key Issues"—A Mounting List of Challenges

Namkung Min-kwan
2026-11-23 05:30:03
[Edaily Namkung Min-kwan SEOK JI-HOEN Sung Gha-hyun Reporter] “While some associate judges are trying to utilize it, it often yields unexpected results, so it’s really just for reference.” (Judge Kim, Civil Division, Seoul Central District Court)

“Compared to generative AI, its ‘limited capabilities’ are even more striking. It’s basically just another search engine on par with those found on portal sites like Naver or Google.” (Judge Lee, Seoul Administrative Court)

Although the Supreme Court’s National Court Administration designated this year as the de facto “first year of the judiciary’s AI era,” it appears that frontline courts have yet to feel any tangible impact.

According to legal circles on the 22nd, the actual utilization of the trial-support AI—which the Court Administration Office completed its first-phase project for early this year and began full-scale operation in February—has remained minimal. While plans are in place to advance the system through phases 2 through 4, critics point out that with the current budget, it is difficult not only to implement improvements that would enhance effectiveness but even to secure the necessary equipment and personnel.

Seoul Court Complex. (Photo = Yonhap News)


Development Schedule Tightened by “0 Won” Budget Execution

The court support AI project, which is being built at a total cost of 16.1 billion won and is scheduled for completion by December 2028, provides a comprehensive collection of materials, including △Supreme Court precedents and judgments △laws and Supreme Court rules △decisions and authoritative interpretations △practical guides and commentaries, among other legal documents. It also aims to incorporate a feature that analyzes user queries to search for relevant legal issues and associated materials, then summarizes and presents the key points.

Judges on the front lines have pointed out that it will be difficult to increase the system’s utility unless fundamental issues—such as search accuracy and response speed—are resolved. Multiple judges who spoke with E-Daily agreed that not only are the accuracy and speed of case law search results lacking, but the system’s capabilities as a generative AI—beyond simple search functions—are significantly inferior to those of commercial AI tools like ChatGPT, Gemini, and Claude, thereby limiting its practicality.

Judge Yoon from the Seoul Central District Court and Judge Ahn from a regional district court both stated, “While the system does produce somewhat relevant search results for lower-court precedents, there are quite a few instances where the answers contain errors or where lower-court precedents unrelated to the issue at hand are included in the search results.” Judge Kim of the Seoul High Court also stated, “Not only is the response time at least three times longer than that of commercial generative AI, but it also lacks functions such as document summarization, issue analysis, and the organization of theoretical grounds or case law, which limits its practical application.”

There is a consensus that securing the necessary budget is the biggest obstacle.

Both inside and outside the courts, the prevailing opinion was that developing a proper AI system to support trials would require a budget of hundreds of billions of won. However, the National Court Administration sought to minimize this cost, requesting 19.078 billion won from the National Assembly. Even so, the actual budget allocated amounted to only 16.1 billion won—about 84% of that amount.

Due to the lack of project funds, even the execution of the secured budget appears to be delayed. Not a single won of the 4.239 billion won allocated for this year’s Phase 2 project had been spent as of the end of last month. Similarly, the sentencing support AI project, which is being pursued separately, has a 0% execution rate for this year’s budget (6.578 billion won).

Experts in the field point out that, considering the costs of graphics processing units (GPUs) and dedicated infrastructure, it will be difficult to move beyond the pilot service phase with the current budget.

Kwon Chang-hwan, a presiding judge at the Ansan Branch of the Suwon District Court and the lead member of the Judiciary’s AI Committee, stated, “The budget for the AI-assisted trial project is extremely small,” adding, “For the approximately 18,000 court personnel—including about 3,000 judges—to use AI services like ChatGPT, we need GPUs capable of handling the number of concurrent users and the scale of the models.” He went on to explain, “Currently, the number of GPUs actually in operation is in the single digits, which is woefully insufficient.”

The budget shortage is also affecting the development environment. Judge Kwon emphasized, “To create a service that staff can use simultaneously, we need not only GPUs and AI models but also a data center,” adding, “At least 400 billion won is required to build AI that is actually usable.”

(Graphic by Reporter Lee Mi-na)


The AI Judges Want Is ‘More Than Just Search’

Given the severe budget shortage, building development infrastructure remains a distant prospect. Judge Kim, a judge at the Seoul Central District Court, pointed out, “I understand that because they cannot secure enough GPUs, they are conducting development, verification, and service operations simultaneously with limited equipment,” adding, “They are also relying on external contractors for development personnel.”

This also acts as a stumbling block to building internal development capabilities within the courts. Judge Kim from the Seoul Western District Court cited the example of the Air Force’s Intelligent Information Systems Management Unit, which built and operated its AI system using internal staff, and suggested that if the budget allows, the courts should also form an internal data management team to gain relevant experience.

Judge Kim went on to mention examples such as: △ the U.S. Administrative Office of the Federal Courts, which last year entered into contracts with legal information providers such as LexisNexis and Thomson Reuters to enable judges and staff to utilize generative AI features; and △ the Singaporean judiciary, which introduced an AI case summary tool for Small Claims Tribunal judges last September and, starting last December, made it available to parties conducting litigation without attorneys. “Our courts need to take note of these international examples, which have improved access to justice not only for the courts but also for citizens,” he said.

The National Court Administration explained, “We are well aware of the feedback from the field regarding accuracy, response speed, and practical functionality. We will implement improvements in stages to provide tangible assistance in daily operations,” adding, “We plan to continue development through April of next year, including adding features such as an integrated summary of first-instance litigation materials and individual case summaries for various litigation documents to the ‘Key Issues Analysis’ menu. “We plan to add features such as support for drafting review reports by around May 2028,” the office explained.

Economy

Corporation

IT·Science

Economy

"Just How Risky Is the Stock Market on Monday?"…Even the Government Is Taking Emergency Measures

In response to the stock market shock originating in the U.S., the government has begun preparations to activate the Securities Market Stabilization Fund. It is also temporarily relaxing the limits on…
2026-11-23 04:50:04

Corporation

"Beyond the Domestic Home Shopping Market to Vietnam": GS Retail Sees Success in Export Consultations with Small and Medium-Sized Enterprises

GS Retail(007070)has supported domestic small and medium-sized partner companies in Vietnam in expanding their overseas sales channels, resulting in export consultation deals totaling approximately $1…
2026-11-22 18:23:06

IT·Science

'Record-Based Social Media' Takes Off in the Age of Short-Form Content… Naver's ThinkBook Surpasses 1 Million Users in North America

In the North American social media (SNS) market dominated by global Big Tech giants such as Instagram and TikTok, Naver (NAVER(035420)) is seeing success with its new initiative. As short-form content…
2026-11-23 05:04:03