Johnson v. Dunn is the case that proved having an AI policy isn't enough if nobody follows it. In July 2025, a federal judge in Alabama disqualified three Butler Snow attorneys from a case and ordered bar regulators notified after they filed motions with hallucinated citations, despite the firm having written AI policies and an Artificial Intelligence Committee. The court called Reeves's conduct "recklessness in the extreme," tantamount to bad faith.


Background

The underlying case was filed in the Northern District of Alabama. Three attorneys from Butler Snow LLP, a well-regarded national firm with over 350 lawyers, represented a party in the litigation. The firm had invested in AI governance. It had written AI policies. It had an Artificial Intelligence Committee. On paper, Butler Snow was doing everything the legal profession recommended.

Attorney Matthew B. Reeves used ChatGPT to generate citations for motions filed with the court. He inserted the citations without verifying a single one. William J. Cranford drafted, signed and filed both motions without reviewing the citations Reeves had added. William R. Lunsford was also involved in the filings.

The gap between the firm's stated policies and the attorneys' actual conduct became the central issue. Butler Snow had the infrastructure for responsible AI use. These attorneys simply didn't use it.

Johnson v. Dunn
No. 2:21-cv-1701-AMM (N.D. Ala. July 23, 2025), 792 F. Supp. 3d 1241
Court
U.S. District Court, Northern District of Alabama
Date
2025-07-23
Category
AI Hallucination / Sanctions
Sanctions
None
AI Case Law: Updated April 2026

What Happened

The court identified hallucinated legal citations in the filed motions. Judge Anna Manasco investigated and found that Reeves had used ChatGPT to generate the citations and inserted them directly into the filings without any verification.

Judge Manasco described Reeves's repeated decisions to repeat AI-generated citations without verifying them as "recklessness in the extreme," tantamount to bad faith. This wasn't a junior associate making a mistake. These were experienced practitioners at a prominent national firm. The court noted that the attorneys' experience level made their failure to verify even more inexcusable.

The investigation also revealed that Butler Snow's AI policies, which existed on paper, had not prevented the misconduct. The firm had an Artificial Intelligence Committee specifically tasked with governing AI use. None of its guidance stopped these attorneys from filing unverified ChatGPT output with a federal court.


The Ruling

Judge Manasco imposed sanctions that went far beyond monetary penalties. All three attorneys, Lunsford, Reeves, and Cranford, were disqualified from representing the client for the remainder of the case. The court directed that the opinion be published in the Federal Supplement, ensuring it would be widely available in legal databases.

The court also ordered the three attorneys to give the clerk a list of every jurisdiction where they are licensed, and directed the clerk to serve the order on the General Counsel of the Alabama State Bar and any other applicable licensing authorities for further proceedings as appropriate. The order also publicly reprimanded them and required them to give a copy to their clients, opposing counsel, the presiding judge in every pending case where they are counsel of record, and every attorney at their firm.

The absence of a monetary fine was deliberate. The court chose sanctions that carried greater professional consequences: removal from the case, published embarrassment, and regulatory notification. For experienced attorneys at a major firm, these consequences are far more damaging than a check to the court registry.

Outcome: Rather than monetary penalties, the court disqualified all three offending attorneys from representing the client for the remainder of the case. The court directed the opinion be published in the Federal Supplement and ordered the clerk to serve the order on the Alabama State Bar and other applicable licensing authorities.

Why This Case Matters

Johnson v. Dunn is arguably the most significant AI sanctions case since Mata v. Avianca. It demolished the assumption that large, well-resourced firms with formal AI governance are safe from AI sanctions. Butler Snow did what the profession recommended: policies, committees, guidelines. It didn't matter because the individual attorneys didn't follow them.

The disqualification sanction was unprecedented in AI cases. Prior courts had imposed fines, CLE requirements, grievance referrals, and license suspensions. Removing attorneys from a case mid-litigation forces the client to retain new counsel, disrupts the entire proceeding, and signals to the legal market that these attorneys can't be trusted with AI-assisted work. The conduct here also has evidentiary implications beyond sanctions: under FRE 901, any AI-generated content submitted as evidence must be authenticated; the same verification failure that caused the sanctions problem creates a separate authentication problem if AI output reaches the record.

The bar notification order amplified the consequences exponentially. Instead of leaving any referral to the opposing side, the court itself sent the order to the Alabama State Bar and other applicable licensing authorities, so discipline is for those authorities to decide.


Lessons for Attorneys

AI policies are necessary but not sufficient. Butler Snow had written policies and a dedicated AI committee, and it still got sanctioned. The lesson for firm leadership is that policies only work if they're enforced, trained on regularly, and embedded into actual workflows. A document sitting in a policy manual doesn't prevent an attorney from pasting ChatGPT output into a brief at 11 PM.

Experience doesn't protect you. The court specifically noted that these were experienced practitioners, and that their seniority made the failure worse, not better. Senior attorneys sometimes assume their judgment compensates for process shortcuts. This case proves otherwise. AI hallucinations don't care about your years of practice or your firm's reputation.

Firms need to build verification into the technology itself, not just the policy manual. If attorneys can use ChatGPT to generate citations and paste them directly into filings without any checkpoint, the workflow is broken regardless of what the policy says. The verification step needs to be structural: a required sign-off, a mandatory database check, a second-reviewer gate. Something that physically prevents unverified AI content from reaching a filing.


The Bottom Line

Johnson v. Dunn proved that AI policies without enforcement are meaningless. Three experienced Butler Snow attorneys were disqualified and reported to bar regulators despite the firm having written AI governance. The policy existed. The attorneys ignored it.

Primary source and what changed

Primary source: Johnson v. Dunn, No. 2:21-cv-1701-AMM, 792 F. Supp. 3d 1241 (N.D. Ala. July 23, 2025) (sanctions order of Judge Anna M. Manasco, read October 29, 2026).

What changed (October 29, 2026): This page was titled "When AI Evidence Went to Court" and described the case as one about AI-generated evidence. That was wrong. The order concerns fabricated case citations that ChatGPT produced for two motions, and the sanctions that followed. We rewrote the title, description, headline and two FAQ answers to say so, corrected the docket number (2:21-cv-1701-AMM), quoted the court as "recklessness in the extreme" about Reeves, and replaced the claim that bar regulators were notified "in every state" with what the order says: service on the Alabama State Bar and other applicable licensing authorities. The order also publicly reprimanded the three lawyers and released Butler Snow LLP without sanction.

AI-Assisted Research. This piece was researched and written with AI assistance, reviewed and edited by Manu Ayala. For deeper takes and the perspective behind the research, follow me on LinkedIn or email me directly.

Price sources, checked 22 October 2026: Gavel. Each figure above is a list price from the vendor's own pricing page, or arithmetic on one. Plans, seat minimums and promotions change; confirm on the vendor's page before you buy.