In May 2026, Seoul Economic Daily reported a case that should reframe how policymakers think about artificial intelligence in the courtroom. A self-represented man surnamed Park, sued by a law firm for 5.1 million won in unpaid fees and separately accused in a criminal complaint by the attorney involved, used generative AI to analyze the filings against him, research the relevant law, and draft his own submissions. The AI flagged concrete defects in the lawyer's own work — incorrect dates in transcripts, and missing dates and signatures on contracts. Park prevailed in the civil suit, and police closed the criminal complaint without referral to prosecutors for insufficient evidence (Seoul Economic Daily).
It is a small case with large implications. Two other lawyers had reportedly declined to represent Park, citing the awkwardness of litigating against a fellow attorney. AI did not just save him a legal bill — it filled a representation gap the market would not. That is the story worth taking seriously, and it deserves a clearer policy response than either triumphalism or panic.
The strongest case for caution
Begin with the regulators' best argument. Courts are not search engines, and a litigant who treats a chatbot as counsel can be badly misled. Generative models still fabricate citations; jurisdictions from the United States to Australia have sanctioned lawyers for filing AI-invented case law. A self-represented litigant has no professional indemnity insurance, no bar discipline, and no duty of candor backstopping their work. If AI lowers the cost of filing, it can also raise the volume of meritless, AI-padded claims that clog dockets and burden opposing parties — a problem courts in several countries are already documenting. And unlike a regulated attorney, an AI tool that gives ruinous advice is, for now, accountable to no one. These are real concerns, not regulatory reflexes.
But Park's case cuts the other way on the question that matters most: it was the licensed professional whose documents contained the errors, and the layperson with AI who caught them. The asymmetry of legal expertise that justifies much of the profession's protected status is exactly what the technology is compressing.
What Korea's new AI law does — and doesn't — reach
South Korea is unusually well placed to think about this. Its Framework Act on AI — the "AI Basic Act" — took effect on January 22, 2026, making Korea one of the first countries with a comprehensive horizontal AI statute (U.S. International Trade Administration). The Act takes a deliberately lighter touch than the EU AI Act: it concentrates binding obligations on "high-impact" systems in sectors such as healthcare, energy, and public services, requires generative-AI providers to disclose when content is AI-generated, and caps administrative fines at roughly 30 million won — about $21,000 — far below European penalty ceilings (Future of Privacy Forum). The Ministry of Science and ICT is running a grace period through 2026 during which fines are generally deferred.
This is, broadly, the right architecture — risk-based, innovation-conscious, and transparency-first rather than prohibition-first (U.S. International Trade Administration). But notice what it does not address. The Act governs AI providers and operators; it says little about an individual citizen using a consumer chatbot to defend themselves in court. Park's use case sits in a regulatory blind spot — neither a "high-impact" deployment nor a professional legal service, just a person and a tool.
The liability question the case actually raises
The instinct in some quarters will be to police AI-assisted self-representation: require disclosure when filings are AI-drafted, or restrict the practice outright. That would be a mistake. The transparency the AI Basic Act already requires — labeling AI-generated output in the market — is a proportionate, light-touch tool. Extending heavy disclosure or liability rules to litigants themselves would penalize precisely the people the justice system underserves.
The more honest reading is that the professional-liability stakes here run toward lawyers, not away from them. If an unrepresented opponent armed with AI can routinely surface dating errors and missing signatures, the standard of care expected of licensed attorneys rises, not falls. That is a healthy market signal. Korea's own data underline why it is arriving fast: the Bank of Korea has estimated a 79% probability that the work of judges, prosecutors, and lawyers could be performed by AI at current technology levels, and lawyer job postings fell more than 18% between 2021 and 2025, from 3,895 to 3,167 (Seoul Economic Daily).
The profession's real anxiety is structural. Seoul National University's Professor Ko Hak-soo has warned of an "AI pyramid" in which senior lawyers capture more value while junior roles — the rung where routine research and drafting once lived — disappear (Korea Times). That is a serious workforce question. It is not, however, a reason to wall ordinary people off from tools that finally let them contest a one-sided fight.
The proportionate path
The right response is not to certify, license, or restrict litigant-side AI. It is to make courts AI-literate: clear guidance that AI-assisted self-representation is permitted, sanctions targeted narrowly at fabricated citations rather than at the use of AI as such, and judicial training so benches can spot both AI errors and AI-surfaced truths. Korea's AI Basic Act got the macro-architecture right by regulating risk, not technology. The Park case is a reminder to apply the same discipline at the micro level — protecting against demonstrable harms like hallucinated authority while preserving the access-to-justice gains that a 5.1-million-won dispute just put on the public record.