Alabama case shows how AI shortcuts can be a fast track to sanctions

We all have read with alarm about attorneys submitting hallucinated citations to the court through their careless use of generative artificial intelligence, leading to sanctions. Most attorneys probably think that could never happen to them. But take heed of the nightmare that befell three attorneys from a prestigious Southern law firm.

In Johnson v. Dunn, No. 2:21-cv-1701 (N.D. Ala., July 23, 2025), the court issued a significant sanctions order after the defendant's attorneys submitted five citations fabricated by ChatGPT in two discovery motions. The court lamented that the reprimands and modest fines commonly issued for AI misuse have not been sufficient deterrents as the problem persists.

Citing fake cases creates many harms. The opposing party wastes time and money exposing the deception, the client may be deprived of legitimate arguments, the court must take time away from other cases to investigate, the reputation of the judges whose names are falsely linked to a fictional opinion can be harmed, cynicism can develop about the justice system, future litigants may claim doubt about the authenticity of an opinion and the public "is justifiably horrified and outraged when filings in a court of law substitute lazy, convenient fictions for the truth." (Id., at p. 24.)

The three sanctioned attorneys played different roles in submitting the motion. Of counsel attorney William Cranford drafted the initial versions of the motions and submitted them to partner Matthew Reeves for review. Reeves inserted cases he found on ChatGPT in support of general propositions. He admittedly failed to verify the citations independently. When Cranford received the revised version from Reeves, he did not check the citations. He signed both motions. Partner William Lunsford is the principal responsible attorney for the client, the Alabama Department of Corrections. His name appeared in the signature block. He reviewed one of the motions after Reeves added the citations but did not independently review their accuracy.

The court analyzed various options for imposing sanctions and decided to rely on the court's inherent authority to discipline attorneys for ethical violations, noting a finding of "subjective bad faith or something tantamount to it is necessary" to support such a sanction. Significantly, even where there is no direct evidence of subjective bad faith, when an attorney's conduct is so egregious it could only be committed in bad faith, the standard is met. (Id., at p. 30.)

The court had no difficulty in finding Reeves, who inserted the false citations without verification, demonstrated complete disregard for his professional duty of candor and therefore acted in bad faith, particularly when he knew from warnings, policies and expectations of his firm of the risks AI will make things up.

As for Cranford, even though he did not know Reeves had used generative AI, he was the signer of the documents and had a responsibility to check the citations, which would have required little effort. The court explained that attorneys who sign motions must know their failure to ensure the motions are accurate will subject them to risk of sanctions. His misconduct was tantamount to bad faith.

The court engaged in extensive analysis regarding Lunsford, the principal responsible attorney who "simply made no effort whatsoever to verify the contents of the motions for himself (or even to ask someone else to check for him)." (Id., at p. 40.) The court identified several other concerns regarding Lunsford. Of most interest is his troubling explanation that AI would not be needed in his cases as they were often similar in fact and law. Thus, the ordinary practice is to re-use material from other cases, apparently without verification. The court concluded that Lunsford did not require his team to perform or verify legal research for his cases. His utter disregard for the truth of filings bearing his name in the signature block was tantamount to bad faith.

Following a review of the multiple AI-false-statement cases nationwide, the court concluded a fine and public reprimand were insufficient. "If fines and public embarrassment were effective deterrents, there would not be so many cases to cite." (Id., at p. 47.) To rectify egregious misconduct, the court ordered: (1) a public reprimand of the three attorneys with a publication order; (2) their disqualification from the case; and (3) referral to the state bar. The attorneys were directed to provide a copy of the order to their clients, opposing counsel and the presiding judges in every pending case where they are counsel of record.

The 50-page order is well worth reading by all attorneys who file motions or briefs with a court even if they have never used generative AI. It should impress everyone with the necessity to have systems in place, which are actually followed, to verify all citations before submitting them to be filed. If the lessons in this case are not heeded, attorneys risk finding themselves reprimanded, disqualified and referred to the state bar for a lack of effort.