By Thorsten Meyer | November 2, 2026 | ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Picture a judge in Korea opening a brief. It cites a Supreme Court ruling, case number and all, in the court’s own format. The reasoning is unfamiliar, so the judge looks it up. Nothing comes back. The ruling does not exist. Further down the same filing, the judge finds something stranger: lines that read like a chat with an AI assistant, notes on litigation strategy and all, pasted into the document without a single edit.
Each of those details comes from real examples that South Korea’s National Court Administration, the administrative arm of the Supreme Court, published on November 2, 2026, with its first survey on fake citations in court filings. The headline travelled fast: about eight in ten judges have seen it happen. The full picture is more precise than that, and in some ways more uncomfortable.
Key takeaways
- Korea’s court administration surveyed its own judges. 360 judges from courts at all levels answered between October 6 and 16, 2026, and the National Court Administration published the results on November 2.
- About 80% of responding civil, family and administrative judges have seen fake or distorted citations. Among judges who had seen them, 196 (80.7%) named precedents that do not exist and 159 (65.4%) named real precedents cited with false content.
- The "eight in ten" headline needs care. It refers to judges handling civil, family and administrative cases. Across all 360 respondents, the Korean daily Kyunghyang Shinmun reports 243 (67.5%), roughly two in three.
- Lawyers are part of the problem, not only laypeople. 164 judges (69.5%) said the parties themselves had cited fake material, and 136 (57.6%) said licensed lawyers had.
- Judges want rules with teeth. 91.7% called for an institutional response; the most popular options were a legal basis for fines (61.4%), a duty to verify citations (61.1%) and a duty to disclose AI use (60.6%).
- Korea is following a path the US and the UK have already walked. The court administration plans to propose changes to the Civil and Criminal Procedure Acts to allow fines, after courts in New York and London have already sanctioned or referred lawyers to regulators.
Products Worth Considering
Apple Watch SE (2nd Gen) GPS 40mm with Starlight Aluminum Case and Sport Band (S/M)
Amazon Kindle Scribe (64GB) with Premium Pen, Metallic Jade
Bose QuietComfort Ultra Headphones (2nd Gen), Wireless Bluetooth Noise-Cancelling Over-Ear Headphones, Black
As an affiliate, we earn on qualifying purchases.
What exactly the Supreme Court published
I went to the source, the Korean-language press release on the Supreme Court’s website, titled "AI 허위 판례 등 인용 관련 설문조사 결과" ("Results of the survey on citations of AI-generated false precedents and similar material"). It was issued by the court administration’s Judicial Support Office and is four pages long. Everything below comes from that document unless I say otherwise; the translations are mine.
The National Court Administration (법원행정처) asked judges at courts of all levels; 360 answered between October 6 and 16, 2026. The reason, per the release: AI hallucinations are producing citations of false and nonexistent precedents, which "impose an unnecessary burden of verification on the opposing party and can hinder the speedy progress of trials." The release does not say how many judges were invited or how the questions were worded. It measures what judges noticed, not how many filings contained fake material.
The core finding, in the release’s words: "about 80% of responding judges in charge of civil, family and administrative cases answered that they had experienced cases in which false or nonexistent precedents were cited."
By type (multiple answers allowed):
| Type of false citation | Judges | Share |
|---|---|---|
| Precedents that do not exist | 196 | 80.7% |
| Real precedents cited with false content | 159 | 65.4% |
| False or nonexistent statutes | 41 | 16.9% |
| False or nonexistent papers and literature | 10 | 4.1% |
Watch the denominator. The release says each percentage is based on the respondents to that question. 196 out of 243 is 80.7%, and 243 is the number of judges Kyunghyang Shinmun reports as having experienced fake citations: 67.5% of all 360 respondents. The release itself does not print that figure, but the arithmetic matches. So "eight in ten" means eight in ten responding civil, family and administrative judges, and about two in three of everyone who answered. Still striking. Just different.
Who submits the fake citations
This is where the survey gets uncomfortable for the legal profession. When judges were asked who had put the false material in front of them, 164 (69.5%) chose "the party themselves" and 136 (57.6%) chose "a lawyer acting as litigation representative or defense counsel." In the release’s words, false citations were found "not only from the parties themselves but in many cases also in documents submitted by lawyers."
One detail matters for English readers. Yonhap’s English version, republished by The Korea Herald, says "164 judges said they themselves have cited false precedents." The Korean original says the 164 judges named the litigants themselves (당사자 본인) as the source. No judge confessed to anything. It is a translation slip, and exactly the kind of error this story is about: a confident sentence that falls apart when you check the original.
The release lists four kinds of real examples:
- Fake judgments generated with AI that copy the format of a Supreme Court decision.
- Supreme Court case numbers attached to unfamiliar legal reasoning, where the precedent cannot be found when searched.
- Real holdings changed, or cited against the meaning of the actual ruling.
- Text from conversations with an AI, or wording about litigation strategy, left in the filing as it was.
The last one is the most telling. Somebody copied a chatbot’s answer and filed it without noticing the chatbot talking back.
What the judges want
91.7% of respondents said an institutional response is needed, and 64.8% said it is "very necessary." The press release includes a table of the measures judges chose (multiple answers, percentages based on respondents to that question):
| Proposed measure | Judges | Share |
|---|---|---|
| A legal basis for fines | 215 | 61.4% |
| A duty to verify cited precedents | 214 | 61.1% |
| A duty to state that AI was used | 212 | 60.6% |
| Sanctions such as disciplinary referral of lawyers | 194 | 55.4% |
Next, the court administration says it wants a legal basis for fines through amendments to the Civil Procedure Act and the Criminal Procedure Act. It will consult the Ministry of Justice and other agencies and submit its opinion to the National Assembly. It is also considering a notice on "points to note when using AI" on the electronic litigation website. Nothing has been legislated yet.
The release also frames the problem as one of access to justice. False citations, it says, "ultimately delay the people’s right to a trial." And it singles out one group: people who go to court without a lawyer "may have difficulty finding and responding to such errors."
Korea did not start from zero
On February 19, 2026, the court administration announced a "false case number check" on its judicial information disclosure portal, live from February 20. The notice says generative AI tools, naming ChatGPT and Claude, learn the pattern of Korean case numbers such as "2023다12345" and combine numbers at random, and that users tend to cite them without doubt when the content looks plausible. Similar cases, it says, "have recently kept occurring" in Korea.
On February 24, 2026, it published an "AI Guidebook for Judges" ("법관을 위한 AI 가이드북"), written by a task force of eight judges and a secretary, with a risk checklist for hallucination, bias and privacy and around 20 worked examples.
So: first a tool to catch fake case numbers, then guidance for judges’ own use, now numbers and a push for sanctions. The February measures were clearly not the end of the problem.
The cases that set the pattern
Mata v. Avianca (New York, 2023). In a personal injury case against an airline, lawyers filed a brief citing judicial opinions that did not exist. On June 22, 2023, US District Judge P. Kevin Castel of the Southern District of New York wrote that "there is nothing inherently improper about using a reliable artificial intelligence tool for assistance," but that "existing rules impose a gatekeeping role on attorneys to ensure the accuracy of their filings." The court found the lawyers had submitted "non-existent judicial opinions with fake quotes and citations created by the artificial intelligence tool ChatGPT," then kept standing by them after the court questioned their existence. It imposed a $5,000 penalty, jointly and severally, on two lawyers and their firm.
His list of harms reads like a preview of the Korean release: "The opposing party wastes time and money in exposing the deception. The Court’s time is taken from other important endeavors."
Ayinde and Al-Haroun (London, 2025). On June 6, 2025, a Divisional Court of the High Court of England and Wales, led by Dame Victoria Sharp, ruled on two cases together. In Ayinde, a housing case against a London council, the grounds cited five fake cases. In Al-Haroun, a claim for £89.4 million in damages against Qatar National Bank, the court found that of 45 citations put before it, "in eighteen instances, the case cited does not exist." The client said he had generated the citations with publicly available AI tools, legal search engines and online sources; his solicitor admitted relying on that research without checking it. One of the fake authorities cited to the judge, Mrs Justice Dias, was a decision attributed to Mrs Justice Dias herself.
The court decided against contempt proceedings and instead referred lawyers to their regulators. It was explicit that this leniency for the junior barrister in Ayinde "is not a precedent. Lawyers who do not comply with their professional obligations in this respect risk severe sanction." It also said freely available generative AI tools "are not capable of conducting reliable legal research" and called on heads of chambers and managing partners to take "practical and effective measures."
How big is this worldwide? No official count exists. A public database kept by researcher Damien Charlotin, which tracks court decisions addressing hallucinated material, listed 2,246 cases when I checked it on November 2, 2026, including entries from South Korea. By its own description it does not cover the wider universe of fake citations nobody wrote a decision about.
How courts elsewhere are responding
The UK judiciary’s guidance for judicial office holders, in its October 31, 2025 version, tells judges that AI tools may "make up fictitious cases, citations or quotes, or refer to legislation, articles or legal texts that do not exist." Lawyers are responsible for what they file; used responsibly, AI need not be disclosed by default, but judges may ask lawyers to confirm they checked AI-assisted research.
On self-represented people, the UK guidance is notably humane. "AI chatbots are now being used by unrepresented litigants. They may be the only source of advice or assistance some litigants receive." Such litigants "rarely have the skills independently to verify legal information" and "may not be aware that they are prone to error." Judges are told to ask about AI use and explain that litigants are responsible for what they file.
That is the tension Korea now has to resolve. A fine deters a law firm. For someone who cannot afford a lawyer and asked a chatbot for help, it is a different thing, and in the Korean survey the parties themselves were named by more judges than lawyers were.
What this means for companies
Law firms first, but not only them. Every company that files documents with a court, a regulator or a tribunal is in the same position.
- The liability sits with whoever signs. In New York, the sanction hit the lawyers and the firm. In London, the court addressed managing partners and heads of chambers directly, and said "a lawyer is not entitled to rely on their lay client for the accuracy of citations." In neither court was "the tool made it up" accepted as a defence.
- Verification becomes a process question. If six in ten responding judges want a duty to verify citations, firms can expect to be asked how they verify, including work done by juniors, contractors or the client.
- Disclosure may become mandatory in Korea. 60.6% of responding judges favour a duty to state AI use in filings. Nothing has been decided, but firms working in Korea should watch the amendments the court administration takes to the National Assembly.
- Your opponent’s fake citation costs you money too. Both Judge Castel and the Korean release point to the opposing party who has to check and rebut invented authorities. That is billable time, and delay, that nobody planned for.
- Legal AI vendors face a trust test. Courts are not banning AI; the UK guidance and the Mata opinion both accept that reliable tools have a place. Tools that link every citation to a verifiable source address exactly what judges complain about.
What this means for people going to court
If you have ever thought about handling a small dispute yourself, with a chatbot as your adviser, this survey is about you.
- Chatbots invent case law convincingly. Korea’s court says AI tools learn what a case number looks like and assemble new ones, sometimes in the Supreme Court’s own layout.
- The court will notice, and you carry the responsibility. The UK guidance tells judges to inform litigants that they are responsible for what they file, and Korea’s court administration says lawyers and parties need to check the accuracy of what they cite. A fake precedent will not help your case and may damage your credibility with the judge.
- There are free ways to check in Korea. Since February 20, 2026, the Supreme Court’s judicial information portal offers a check of whether a case number exists. That does not tell you whether the case says what a chatbot claims, but it catches the most blatant inventions.
- You may be on the receiving end. If the other side files invented case law, you have to spot it. The Korean release names people without lawyers as the group that struggles most with that.
- Courts may soon spell out the rules. Korea is considering a notice on AI use on its e-litigation website.
None of this is legal advice. If your case matters, a qualified lawyer, who checks their sources, is still the safest route.
What we don’t know yet
- The method. The release does not say how many judges were invited or how questions were worded. Judges who had seen fake citations may have been more likely to answer.
- How often, not just whether. The survey counts judges who saw at least one case, not how many filings are affected or whether it is rising.
- Whether every fake citation came from AI. The survey is framed around AI hallucinations, and the examples clearly involve AI. But a fake or misquoted precedent can also come from careless copying or deliberate manipulation. The release does not separate these.
- The criminal side. The headline finding covers civil, family and administrative judges. The release does not give a comparable figure for criminal judges, though it proposes amending the Criminal Procedure Act as well.
- What the law will look like. None of the favoured measures is law yet, and the release includes no response from the Ministry of Justice or the National Assembly.
- The overall share. The 67.5% figure comes from one outlet, Kyunghyang Shinmun. It fits the release’s numbers, but the release does not print it.
The bottom line
The Korean survey does not prove that AI is wrecking courtrooms. It proves something narrower and more useful: in Korea, a clear majority of the civil, family and administrative judges who answered have already caught invented or distorted authorities in filings, and more than half of the judges who answered the question about the source named licensed lawyers among those responsible.
Three years after Mata v. Avianca, the lesson has not changed. Using AI is not the failure. Filing what it produces without checking is. What is changing is the price. New York imposed a fine. London referred lawyers to regulators and warned of "severe sanction." Korea’s judges now want fines, verification duties and disclosure written into procedural law.
For companies, that means building verification into the workflow before a regulator or a judge does it for you. For ordinary people, it means treating a chatbot’s case citations as a lead to check, never as an answer. The case number might look perfect. The court will still look it up.
Sources
- National Court Administration, Supreme Court of Korea, "AI 허위 판례 등 인용 관련 설문조사 결과" (Results of the survey on citations of AI-generated false precedents), press release, November 2, 2026, Korean original read, scourt.go.kr
- National Court Administration, Supreme Court of Korea, "사법정보공개포털 ‘허위 사건번호 확인’ 서비스 제공 알림" (Notice of the false case number check on the judicial information portal), February 19, 2026, scourt.go.kr
- National Court Administration, Supreme Court of Korea, "「법관을 위한 AI 가이드북」 발간" (Publication of the AI Guidebook for Judges), February 24, 2026, scourt.go.kr
- Yonhap News Agency, "About 80 pct of Korean judges encounter AI-generated fake legal precedents: survey", November 2, 2026, en.yna.co.kr
- Kyunghyang Shinmun, "AI와 대화 내용이 서면에 그대로…민사 법관 80%가 ‘허위 판례 경험’", November 2, 2026, khan.co.kr
- Law Times (Beopryul Sinmun), "판사 80% ‘AI 허위 판례 담긴 서면 받아봤다’", November 2, 2026, lawtimes.co.kr
- US District Court for the Southern District of New York, Judge P. Kevin Castel, "Mata v. Avianca, Inc., 22-cv-1461 (PKC), Opinion and Order on Sanctions", June 22, 2023, storage.courtlistener.com
- High Court of England and Wales, Divisional Court, "Ayinde v London Borough of Haringey, and Al-Haroun v Qatar National Bank, 2025 EWHC 1383 (Admin)", June 6, 2025, judiciary.uk
- Courts and Tribunals Judiciary, "Artificial Intelligence (AI) Guidance for Judicial Office Holders", October 31, 2025, judiciary.uk
- Damien Charlotin, "AI Hallucination Cases" database, retrieved November 2, 2026, damiencharlotin.com
This article is for information only. It is not legal advice. If you are involved in court proceedings, consult a qualified lawyer in your jurisdiction.
