Artificial intelligence is transforming the legal sector around the world, streamlining work for legal professionals and driving innovation in an otherwise document-heavy sector. In South Korea, however, this progress is being stifled by a lack of open access to court decisions and overly restrictive practices when it comes to disclosing court decisions.
Artificial intelligence is transforming the legal sector around the world, streamlining work for legal professionals and driving innovation in an otherwise document-heavy sector. In South Korea, however, this progress is being stifled by a lack of open access to court decisions and overly restrictive practices when it comes to disclosing court decisions.A local legal AI startup recently provided a startling figure — only 0.08 percent of court decisions are publicly disclosed. The rest are either sealed or heavily redacted when they are made publicly available.
The country’s constitution guarantees the principle of open justice. Article 109 explicitly states that court trials and rulings are open to the public. However, this is far from a reality. In practice, the courts take an overly cautious approach to privacy and confidentiality, resulting in extremely limited access to legal documents. This lack of transparency is becoming a major obstacle to the development of legal AI.
With little access to court rulings, legal AI developers are starved of the high-quality data required to train their models. The closed nature of South Korean courts risks leaving local legal AI development fall behind others in countries such as Australia and the US where legal AI services thrive on the back of the vast troves of open legal data.
— Restrictive approach to court documents —
In South Korea, court decisions are only disclosed to the parties involved in a case. Others can apply to obtain digital copies through the court website for a small fee of 1,000 won (about $0.60), but the progress is slow and the result is often disappointing.
Before release, court staff remove personal information such as names of individuals, companies, institutions and even geological locations and replace them with generic letters like A, B, or C. As a result, the published version of a decision is so stripped of detail that it’s difficult for anyone without prior knowledge of the case to understand.
Gene Lee, CEO of legal AI startup LBox and a lawyer, displayed one such decision at a parliamentary conference last week in Seoul. It was riddled with generic letters, offering little usable information.
“Excessive anonymization undermines the value of court decisions as data for AI,” Lee said. “It also undermines legal transparency in the South Korean legal system.”
In many cases, anonymized court decisions are not even available if litigants apply for access restriction to make entire decisions withheld from public access and viewing. This happens frequently in antitrust appeals and other cases involving foreign companies.
The opacity is not grounded in law — rather, it contradicts with the constitution. Experts say it stems from a fear within the judiciary that court data could be misused somehow. This overly cautious stance mirrors the attitude of other government agencies.
Bae Il-kwon, director of the public AI and data bureau at the Ministry of Interior and Safety, admitted that about 700,000 public data disclosure requests are rejected annually — mostly due to fears about leaking personal information and legal liabilities.
"The provision of such data remains extremely limited due to high costs, the complexity of handling data properly and legal risks,” Bae said at the conference.
— AI needs data to thrive —
The lack of legal data doesn’t just slow down AI development — it undermines the broader legal tech ecosystem. Without access to precedent, legal research becomes slower and more expensive. Moreover, legal AI services require large volumes of past rulings to train models and deliver accurate results. Without it, developers are left with little to work with, and people continue to rely on expensive legal services or personal connections. Legal professionals often rely on unofficial networks to access past rulings for research.
“Without case law, it’s impossible to predict how a case will unfold,” Rachel Boram Won, head of global business, marketing and communication at Bhsn, one of South Korea’s leading legal tech firms, told MLex. She added that it’s not just legal departments that need access to past rulings, other departments rely on them to anticipate and minimize regulatory risks.
“That’s why the number of rulings a platform has becomes a competitive advantage,” she said.
Startups like LBox and Bigcase advertise their collections, boasting databases of up to 4 million decisions. But that accumulation depends on painstaking efforts to acquire rulings, often through manual requests and purchases.
In contrast, in countries with open court data, legal AI services compete based on the quality of their technology — not on how much data they’ve hoarded.
“The race to develop AI is essentially a race against time — a competition to run more experiments, conduct more tests and gather more service experience to refine models and services. In this process, access to data is critical. There’s concern that without sufficient access to legal data, we risk falling behind.” Hong Dae-sik, law professor at Sogang University and a former judge, told MLex.
— Innovation on ice —
South Korean legal tech companies are urging the government and the judiciary to make more court decisions accessible.
“We believe far more rulings should be made public,” said Won of Bhsn. Her company, which also offers contract drafting and review and other legal queries, has had to build and update its case law database manually.
Because the number of published rulings is very limited, the company has no choice but to purchase court decisions or make copies manually. Won said this creates a considerable burden on the company.
Collecting as many court decisions as possible becomes crucial to build a robust reference database to support retrieval-augmented generation, or RAG, to help reduce hallucinations in generative AI by grounding outputs in reliable data. But without rulings, these systems can’t function properly, according to Won.
— Signs of progress, but a long way to go —
Legal AI startups are lobbying for more open access by raising awareness of the issue at parliament and the government. They are raising hopes on recent legislative developments, such as the AI Basic Law, which mandates government support for generating and managing training data for AI.
The AI Basic Law requires the government to devise a plan to promote the generation, accumulation, management and distribution of data for training AI and allocate appropriate budget to support the implementation of the plan.
A proposed revision of the Rules on the Promotion of Access to and Use of Court Public Data could also signal change. It would require the National Court Administration, responsible for the administrative affairs of the judiciary, to develop plans for expanding public access to legal data.
And in a recent decision, the Supreme Court clarified that courts cannot be seen as personal-data processors, and delivering court documents is not the same as providing personal data, which suggests there may be a way to treat court records as public data, according to experts.
Still, change is unlikely to come quickly. Experts cite limited resources and deeply entrenched norms over privacy and confidentiality within the courts as major barriers.
“Courts have their own deidentification processes, and privacy and confidentiality remain their top priority. But they need to improve public engagement and be more open to collaborating with the private sector to make legal data more accessible and usable,” said Professor Hong.
Please e-mail editors@mlex.com to contact the editorial staff regarding this story, or to submit the names of lawyers and advisers.