Lawyers Sanctioned for AI Fake Citations: Cases and Lessons

01 / PRACTICE NOTES

Mata v. Avianca set the template: a $5,000 penalty, letters to the judges and a finding of bad faith.

In Mata v. Avianca, Inc., No. 22-cv-1461 (PKC), 678 F. Supp. 3d 443 (S.D.N.Y. June 22, 2023), Judge P. Kevin Castel sanctioned two lawyers and their law firm for submitting non-existent judicial opinions with fake quotes and citations created by ChatGPT, and for continuing to stand by those opinions after judicial orders called their existence into question. The court found bad faith on the part of the individual respondents.

The sanctions rested on Rule 11 or, alternatively, the court’s inherent authority. Respondents were ordered to send a letter to the plaintiff attaching the opinion and the hearing transcript, to send a letter to each judge falsely identified as the author of a fake opinion, to file copies of those letters and to pay a $5,000 penalty, jointly and severally, into the court’s registry. The opinion lists the harms that follow a fake citation: the opposing party spends time and money exposing it, the court’s time is taken from other matters and the client may be deprived of arguments based on authentic precedent.

The court also drew the line that still frames the issue. It wrote that there is nothing inherently improper about using a reliable artificial intelligence tool for assistance, and that existing rules impose a gatekeeping role on attorneys to ensure the accuracy of their filings.

02 / PRACTICE NOTES

Courts sanction the signer, not the software: Park, Gauthier and Wadsworth.

Park v. Kim, No. 22-2057, 91 F.4th 610 (2d Cir. Jan. 30, 2024) (per curiam), involved a reply brief that cited a case that does not exist. Counsel admitted she had generated it with ChatGPT and, when ordered to produce the decision, wrote that she was unable to furnish a copy. The Second Circuit referred her to its Grievance Panel under Local Rule 46.2 and ordered her to give her client a copy of the ruling, translated into Korean if necessary, and to certify that she had done so. It stated that, at the very least, the duties imposed by Rule 11 require that attorneys read, and thereby confirm the existence and validity of, the legal authorities on which they rely.

In Gauthier v. Goodyear Tire & Rubber Co., No. 1:23-CV-281 (E.D. Tex. Nov. 25, 2024), counsel filed a response citing two cases that do not exist and quoting language that could not be located in the cited authority. He explained that he used a generative AI tool to produce the response and failed to verify it, and that a verification feature in a legal research service did not flag the problems. He attempted to locate the cases only after the court’s show-cause order, although the opposing party had flagged them in its reply. The court imposed a $2,000 penalty, a one-hour continuing legal education course on generative AI and delivery of the order to his client with a certification on the docket.

Wadsworth v. Walmart Inc., 348 F.R.D. 489 (D. Wyo. Feb. 24, 2025), shows how responsibility spreads across a team. Motions in limine cited nine cases, and eight did not exist. The drafting attorney’s pro hac vice admission was revoked and he was fined $3,000. The supervising attorney, who was not the drafter, and the local counsel who sponsored the pro hac vice admissions were each fined $1,000. The court stated that the supervising attorney had a nondelegable duty to ensure a motion is supported by existing law, and it credited the drafter’s candor as a mitigating fact.

03 / PRACTICE NOTES

Sanctions escalate when counsel repeat the error or defend it: Johnson v. Dunn and Coomer v. Lindell.

In Johnson v. Dunn, No. 2:21-cv-1701-AMM, Doc. 204 (N.D. Ala. July 23, 2025), three attorneys for the defendant confirmed in writing and at a hearing that citations in two motions were hallucinations of ChatGPT. Judge Anna Manasco found a fine and a public reprimand insufficient. She publicly reprimanded the three attorneys, ordered them to give a copy of the order to their clients, opposing counsel and the presiding judge in every pending case and to every attorney in their firm, directed publication of the order in the Federal Supplement, disqualified them from further participation in the case and referred the matter to the Alabama State Bar and other licensing authorities. The court noted that they had repeated warnings, internal controls, firm policies and access to major legal research databases, and it released the firm and two other attorneys from the proceedings without sanction.

In Coomer v. Lindell, No. 22-cv-01129 (D. Colo.), Judge Nina Y. Wang sanctioned counsel in 2025 after a response brief contained nearly thirty defective citations, including cases that do not exist, and counsel admitted using AI. The court imposed $3,000 on lead counsel and his firm and $3,000 on co-counsel, finding that they failed to meet the reasonable level of competence and care that Rule 11 expects. After trial, a further brief described a district court decision as a Tenth Circuit decision and cited it for a proposition outside its holding. On May 7, 2026 the court made its show-cause order absolute as to lead counsel and sanctioned him and his firm another $5,000, jointly and severally, while discharging the order as to co-counsel. It shared other courts’ concern that relatively modest monetary sanctions are insufficient to deter attorneys from haphazardly using AI tools to craft legal arguments.

04 / PRACTICE NOTES

The orders share one pattern: the lawyer who signs owns the verification.

Across six courts, the reasoning is consistent. Rule 11(b)(2) certifies that legal contentions are warranted by existing law after a reasonable inquiry, and a decision that does not exist is not existing law. The courts did not treat the choice of tool as the wrong; Mata says reliable AI assistance is proper. They sanctioned the missing step between the tool’s output and the signature.

Three aggravating facts recur. Counsel failed to open the cited source, counsel failed to correct the filing when opposing counsel or the court raised the problem, and counsel delegated the check without confirming it happened. The mitigating facts recur as well: prompt candor with the court, withdrawal or correction of the filing and a record of the review that was done.

05 / PRACTICE NOTES

The pre-signature checklist that would have prevented each of these orders.

The steps below map to the failures in the six orders. They complement the citation-ledger method in the guide on verifying case citations before you file.

1. Open every authority at its source before the paper leaves the office; read the cited passage and the reasoning around it.
2. Match the proposition to the holding, the court and its level: a district court decision is not circuit authority.
3. Compare every quotation word for word with the original opinion.
4. Verify with the source itself, since another AI answer can repeat the same error.
5. Name the owner of the cite-check for each filing and confirm the owner finished; supervising and local counsel who sign review it too.
6. Re-check after final edits, so the filed version is the verified version.
7. Withdraw or correct promptly when opposing counsel or the court questions a citation, and tell the court what happened.
8. Save the ledger and the review date with the matter.

06 / PRACTICE NOTES

Work with the sources open and the review saved in the case.

RuleX is built around that checklist. It researches federal and state law, presents the authority with sources you open from the answer, works with up to twenty documents at once while preserving citations and context and writes the draft in Word beside the file. The case log and tasks hold the cite-check assignment and the date it was completed. Start the 7-day free trial, add a real matter and open the sources from the answer.