이 페이지에서4분 읽기
개요
Avianca in 2023, where a federal judge imposed a $5,000 penalty on two attorneys and their firm. These cases matter because the failure is skipped verification, not use of the tool, and courts treat unchecked citations as a breach of the lawyer's duty to the court.
심층 분석
Mata v. Avianca, Inc. was an ordinary personal injury suit in the Southern District of New York: a passenger said a metal serving cart struck his knee on a 2019 flight. When Avianca moved to dismiss, the plaintiff's opposition cited decisions, including one styled Varghese v. China Southern Airlines, that did not exist. Neither opposing counsel nor the court could find them. The lawyer who did the research, Steven Schwartz, had used ChatGPT, believed it worked like a search engine, and asked it to confirm the cases were real, which it did. His colleague Peter LoDuca signed the filings. When the court ordered them to produce the decisions, they submitted AI-generated text presented as opinions. On June 22, 2023, Judge P. Kevin Castel imposed a $5,000 penalty on the two lawyers and their firm under Rule 11 and ordered them to write to their client and to each real judge falsely named as an author. The court made clear that using AI was not the problem. The lawyers failed by abandoning their responsibilities and continuing to stand by the fake cases after they were questioned. Later cases show the pattern persisting. The Second Circuit in Park v. Kim (2024) referred an attorney for possible discipline. In United States v. Cohen, Bard-generated citations reached a filing, though the judge declined sanctions. In Wadsworth v. Walmart (2025), lawyers were sanctioned over citations from their firm's in-house AI tool, showing that internal or legal-specific tools are not immune. Some judges, including Judge Brantley Starr of the Northern District of Texas in 2023, now require certifications about AI use. The central misconception is that asking the AI to double-check itself counts as verification. It does not. A model asked whether a case exists will often generate a confident yes. Verification means retrieving the actual opinion from an authoritative source and reading it.
전략적 영향
위험과 안전
치명적인 AI 피해와 일상적인 AI 피해는 누가 위험을 이해하고 누가 조치를 취할 수 있는지에 따라 달라집니다.
더 명확한 결정들
공공 및 전문 지식은 강력한 안전 정책이 정치적으로 가능한지 여부를 결정합니다.
과장된 과장을 뚫고 나가기
명확한 설명은 과대광고, 연구실 홍보, 모호한 윤리 연극에 의한 포착을 줄입니다.
The Future of Lawyers Sanctioned for AI-Fabricated Case Citations
Courts continue to encounter AI-fabricated citations, from experienced lawyers at large firms as well as from self-represented litigants, so the problem is not fading on its own. Responses vary: some judges issue standing orders requiring disclosure or certification, others rely on existing Rule 11 duties, and bar regulators increasingly treat repeated or concealed failures as disciplinary matters. Citation-checking features in research and drafting platforms may catch more errors before filing, but confirming that a case exists is easier than confirming it supports the stated proposition. The durable fix remains a firm-wide habit: every authority is read before it is cited.
실제 구현
In Mata v. Avianca, an attorney used ChatGPT to research an airline injury case, then asked ChatGPT whether the cases were real; it said they were, and the fabricated authorities were filed with the court.
In Park v. Kim (2024), the Second Circuit referred a lawyer to its grievance panel after a reply brief cited a nonexistent decision that she acknowledged had come from ChatGPT.
In United States v. Cohen, citations that Michael Cohen found using Google Bard and passed to his lawyer ended up in a court filing; Judge Jesse Furman declined to impose sanctions but found the citations had not been checked.
In Wadsworth v. Walmart (2025), a federal court in Wyoming sanctioned attorneys after motions cited cases produced by their firm's internal AI tool, with the heaviest penalty on the lawyer who drafted the filing.
위험 및 가드레일
실존적 위험을 공상과학처럼 다루면서 능력을 합성합니다.
높은 자율성 하에서 정렬과 표면 제품 안전성을 혼동합니다.
영어가 아니거나 전문가가 아닌 청중에게는 품질이 낮은 소스만 남겨 둡니다.
구현 로드맵
제품 손상, 오용, 통제력 상실/잘못 정렬 위험을 분리합니다.
일정과 심각도에 대한 귀하의 견해를 바꿀 수 있는 증거가 무엇인지 물어보십시오.
마케팅 주장보다 기본 소스와 구체적인 평가를 선호하세요.
인식뿐만 아니라 경력, 정책, 자금 조달 또는 기술 등 하나의 행동 경로를 식별하십시오.
계속 탐색하세요
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the Lawyers Sanctioned for AI-Fabricated Case Citations quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
자주 묻는 질문
What is Lawyers Sanctioned for AI-Fabricated Case Citations?
Lawyers have been sanctioned for filing briefs containing case citations that generative AI tools invented, most famously in Mata v. Avianca in 2023, where a federal judge imposed a $5,000 penalty on two attorneys and their firm. These cases matter because the failure is skipped verification, not use of the tool, and courts treat unchecked citations as a breach of the lawyer's duty to the court.
In Mata v. Avianca, what happened when the researching attorney asked ChatGPT whether the cited cases were real?
ChatGPT confirmed the invented cases were real, which illustrates why asking a model to verify its own output is not verification.
What sanction did Judge Castel impose in Mata v. Avianca on June 22, 2023?
The court imposed a $5,000 penalty under Rule 11 and required the lawyers to notify their client and each judge falsely identified as an author of a fake opinion.
According to the guide, what did the court in Mata identify as the lawyers' real failure?
The court said AI use itself was not the problem. The failure was not verifying and then continuing to defend the fabricated authorities.
What lesson does the guide draw from Wadsworth v. Walmart?
The citations came from the firm's own AI platform, showing that in-house or legal-specific tools do not remove the need for verification.
Why can a language model produce a fake citation that looks convincing?
The predictable structure of party names, reporter, volume, page, court and year lets a model generate plausible-looking strings without any link to a real opinion.
계속 학습하세요
관련 가이드
이 주제에 대해 선택된 추가 가이드