Independently reportedLive since May 2025Added Oct 2026

Butler Snow

Butler Snow partner's unchecked ChatGPT research leads to fake citations and three lawyers disqualified

In May 2025 two court filings by Butler Snow, a US law firm, contained five citations that a partner had obtained from ChatGPT and not verified. In July 2025 a federal judge reprimanded and disqualified three of its lawyers from the case, referred them to the state bar, and released the firm itself without sanction.

Key figures

Copied exactly from the sources. Copy any figure with its citation.

five

Problematic citations found across two motions filed in May 20251

more than 2,400 separate legal citations across 330 filings

Citations checked in an independent review commissioned by Butler Snow of its other filings in the same courts1

What it took

CostNot disclosed
TeamNot disclosed
TimelineNot disclosed

Cost, team size and timeline were not disclosed in the sources.

The problem

Lawyers who sign a court filing certify that the law they cite is real and says what they claim. Butler Snow was defending the former commissioner of the Alabama Department of Corrections in an inmate's lawsuit. In May 2025 the inmate's side pointed out that two of its motions contained citations that appeared to be made up.

What they built

Butler Snow did not build a system for this. A partner, Matthew Reeves, used ChatGPT on his own to find case law supporting two arguments. The firm had warned its lawyers about large language model risks in June 2023 and required written permission from a practice group leader to use such tools for research. It had given lawyers a different, approved tool, Westlaw's CoCounsel, in January 2025.

How they did it

Mr. Reeves told the court that he searched ChatGPT for supporting cases, did not check them in Westlaw or the court system before inserting them, and violated the firm's policy. Other lawyers signed and filed the motions. In all, the court counted five problematic citations. Some were to cases that do not exist, and others pointed to real but unrelated cases.

What went wrong

The judge could not find the cases and on May 16, 2025 ordered the lawyers and the firm to explain themselves. The firm apologised, called the lapse inexcusable, hired Morgan Lewis to check its other filings, and promised new training and a pre-filing citation check. Morgan Lewis, using 28 attorneys, reviewed more than 2,400 separate legal citations across 330 filings and found no other fabricated ones, according to the order.

Results

On July 23, 2025 the judge publicly reprimanded Mr. Reeves, William Cranford and William Lunsford, disqualified them from the case, ordered the order published, and referred the matter to the Alabama State Bar. She said the reprimands and small fines common in such cases had not been enough to deter this behaviour. Citing the independent review, she did not suspend them from practice, and she released the firm and two other lawyers without sanction.

Lessons for your project

  1. A written AI policy is not a control. The partner knew the firm required written approval to use ChatGPT for research and did not follow it. Pair policy with technical blocks and approved tools.

  2. Make every signer responsible. The court sanctioned all three attorneys named on the filings even though one partner supplied the citations, because signing certifies the law was checked.

  3. Verify every citation against an authoritative database before filing, regardless of who added it or which tool produced it.

  4. Investigate beyond the single incident. The firm hired an outside firm to re-check its other filings. The judge cited that review as a reason not to suspend the three lawyers, and released the firm because she found no bad faith or recklessness by it.

  5. Plan for the client impact. Disqualification removed the firm's lawyers from a live case, and the court also ordered the three lawyers to give a copy of its order to their clients in every pending case.

Caveats

  • ChatGPT was not an approved deployment at the firm. The record shows a partner used it contrary to firm policy, while the firm had approved a different tool, Westlaw CoCounsel, in January 2025. This entry records the incident and the firm's response, not a sanctioned rollout.
  • No model version is given. The sources name only ChatGPT.
  • The deployed date shown is the date of the first problematic filing. The partner told the court he had made limited use of various AI products since about March 2024.
  • The sanctions are the court's findings. The ABA Journal reports the firm's own account of an inexcusable lapse and its pledged new training and protocols.
  • The firm said the State of Alabama would bear no cost for the errors or the remediation.

Sources

  1. Sanctions Order, Johnson v. Dunn, No. 2:21-cv-1701-AMM (N.D. Ala.), Doc. 204

    United States District Court for the Northern District of Alabama (via CourtListener) · 2025-07-23 · Filing · storage.courtlistener.com

Summarized in our own words from the sources above. Spotted an error? See corrections.