Avoiding Sanctions in the Gen AI Era: Practical Guardrails for Lawyers
4.21.2026
Generative artificial intelligence has now been part of the public legal landscape for nearly three years, and one danger has consistently eclipsed the rest: hallucinations.[1] In this context, “hallucinations” are outputs that present false information as fact. They can stem from a range of causes, including gaps or errors in training data, imprecise or confusing prompts, or responses that drift away from the question being asked.[2] In legal practice, the consequences are both familiar and serious: nonexistent cases, inaccurate or invented quotations, tendentious readings of actual authority and even wholly fabricated legal rules.[3]Courts across the United States continue to encounter AI-driven filing errors with notable frequency, particularly in the form of bogus citations. In many of these matters, the result has been sanctions or other disciplinary consequences for counsel.
Yet these outcomes are not inevitable. In most instances, they are preventable with a basic understanding of how generative AI tools generate answers and a disciplined approach to verification. The obligation is not new: Attorneys have always been required to confirm that the authorities and assertions in their filings are accurate before presenting them to a tribunal.
Against that backdrop, Deutsche Bank Natl. Trust Co. v. LeTennier[4] illustrates the practical overlap between generative AI misuse, and the standards courts apply when determining whether an attorney or party has engaged in frivolous conduct warranting sanctions.
GenAI Misuse and Its Consequence
The underlying dispute began with a 2006 loan evidenced by a note and secured by a mortgage on Delaware County property. After the defendant defaulted, the plaintiff commenced the action in March 2018 and ultimately obtained summary judgment, which the Appellate Division affirmed. The defendant then filed repeated post-judgment motions to renew, reargue, or vacate, resulting in Supreme Court orders in March 2023, August 2024, and September 2024 denying relief and, in the later orders, imposing motion restrictions and awarding costs and legal fees based on frivolous, vexatious motion practice.
On appeal from those orders, the defendant’s filings became the focus. His opening brief cited six nonexistent cases, prompting the plaintiff to seek sanctions and to argue the citations were AI-generated. The defendant initially attributed the problem to citation or formatting mistakes, but later submissions included additional fabricated cases and unreliable characterizations of actual authority. Although he acknowledged the seriousness of citing fictitious cases, he offered no meaningful corrections. Counsel ultimately conceded that AI had been used in preparing the papers and claimed they were checked, but the record reflected otherwise. Across five appellate filings, the defendant cited at least 23 fabricated cases and made other material misstatements of fact or law.
The court characterized the use of fabricated authority as inherently meritless and therefore frivolous under 22 N.Y.C.R.R. 130-1.1(c)(1), and emphasized that an attorney’s signature on a filing certifies, after a reasonable inquiry, that the contentions are not frivolous.[5]
The explanations offered for the false citations were deemed not credible, and counsel’s suggestion that most citations were accurate was still described as incompatible with basic duties of candor.[6] Most troubling, more than half of the fabricated cases were submitted after defendant was already on notice of the problem, with the misconduct escalating rather than stopping. Instead of correcting course, counsel failed to take meaningful remedial steps and effectively minimized the significance of the fake authorities at oral argument.
Deterrence and Punishment: The Court’s Rationale for Sanctions
When selecting an appropriate sanction, courts emphasize that sanctions are not only punitive (addressing past misconduct) but also deterrent, aimed at preventing future frivolous conduct by the litigants involved and the bar more broadly.[7]
Courts nationwide have responded to AI-hallucinated citations and false legal propositions with a wide range of sanctions, from warnings and nominal penalties to awards in tens of thousands of dollars. The sanction level typically turns on factors such as the number of fabricated cases or propositions, whether false quotations were used, whether the filer continued submitting fabricated authorities after being put on notice, whether the error was promptly admitted and corrected, the presence (or absence) of remorse and the practical impact on the proceeding. In matters involving comparable volumes of fabricated citations across multiple filings, courts have imposed significant monetary sanctions.[8]
Because this was New York’s first appellate-level decision addressing sanctions for generative AI misuse, the court held that a $5,000 monetary sanction against defense counsel was warranted to punish the conduct and deter similar behavior by both the defendant and the broader bar.
The court emphasized that generative AI is not categorically prohibited in preparing court submissions; the problem arises when lawyers rely on it without adequate training and meaningful human oversight. Using it does not lessen the duty to verify facts and citations: Failure to do so may be sanctionable depending on the circumstances.
The court agreed with the plaintiff that the appeal itself was frivolous. It pointed to the defendant’s repeated motion practice, including at least seven motions attacking the summary judgment and foreclosure rulings and largely rehashing standing arguments already rejected, along with escalating conduct that drew multiple warnings, the filing of more than 20 irrelevant documents that were struck, and continued pro se filings even while represented.
The court viewed the AI-related misconduct as a further escalation in a pattern of delay and harassment, making the appeal a continuation of protracted, frivolous litigation despite prior sanctions and warnings. Citing indications that some appellate submissions may have originated with the defendant rather than counsel, and emphasizing deterrence and the conservation of judicial resources, the court imposed additional sanctions of $2,500 on the defense counsel and $2,500 on the defendant for pursuing the appeal, and concluded that the remaining arguments were meritless, academic or frivolous.
Conclusion
LeTennier reinforces a bright line lesson: Generative AI can be a helpful drafting and research aid, but it does not, and cannot, displace the lawyer’s nondelegable duties of competence, supervision and candor to the tribunal.[9]
As the American Bar Association has emphasized, lawyers must understand AI’s capabilities and limitations and independently verify outputs before relying on them in client work or court submissions, precisely because hallucinations can produce plausible looking but false “authority.”[10]
Recent sanctions decisions show courts are increasingly willing to treat hallucinated citations and misrepresented holdings as sanctionable conduct under Rule 11 and/or inherent authority, particularly where the errors reflect a lack of reasonable inquiry or persist after notice.
The takeaway for practitioners is operational: Treat generative AI output as unverified work product, require a primary source pull for every cited case, perform quote/record checks and implement a documented supervisory sign-off, especially for filings and appellate briefs where accuracy and candor are paramount.
It also matters how counsel reacts once a problem is identified. Prompt corrections, transparency, and meaningful remedial steps tend to mitigate consequences, while minimization or repetition after warning tends to aggravate them.
Finally, because courts are rapidly adopting standing orders and local rules addressing generative AI use, including disclosure or certification requirements, lawyers should track jurisdiction-specific requirements and build compliance into their filing workflow.[11]
This article appears in a forthcoming issue of NYLitigator, the publication of the Commercial and Federal Litigation Section. For more information, please see nysba.org/comfed.
Lorenzo Rovini, LL.M., is a dual-qualified attorney (Italy and New York) and an associate at Paykin Law, where he focuses on real estate and complex commercial litigation. He is a member of the Chartered Institute of Arbitrators, as well as the NYSBA Real Property Law, Commercial and Federal Litigation, and Business Law sections.
Endnotes:
[1] Zach Warren, GenAI Hallucinations Are Still Pervasive in Legal Filings, but Better Lawyering Is the Cure, Thomson Reuters, Aug. 18, 2025, https://www.thomsonreuters.com/en-us/posts/technology/genai-hallucinations/.
[2] See id.
[3] See Brandon M. Fierro, Short Circuit Court: AI Hallucinations in Legal Filings and How To Avoid Making Headlines, Cole Shotz P.C., Aug. 4, 2025, https://www.coleschotz.com/news-and-publications/short-circuit-court-ai-hallucinations-in-legal-filings-and-how-to-avoid-making-headlines/.
[4] See Deutsche Bank Natl. Trust Co. v. LeTennier, 2026 NY Slip Op. 40 (3rd Dep’t, 2026).
[5] See 22 N.Y.C.R.R. 130-1.1a(b)(1).
[6] See Rules of Professional Conduct, 22 N.Y.C.R.R. 1200.0, rule 3.3(a)(1).
[7] See Levy v. Carol Mgt. Corp., 260 A.D.2d 27, 34 (1st Dep’t, 1999).
[8] See, e.g., Noland v. Land of the Free, L.P., 114 Cal App 5th 426, 449, 336 Cal Rptr 3d 897, 912 (Cal Ct App 2025) (involving 21 fabricated case citations and quotations by attorney; total sanction $10,000); Mid Cent. Operating Engrs. Health & Welfare Fund v. HoosierVac LLC, 2025 WL 1511211, *2, 2025 US Dist LEXIS 100748, *3-4 (SD Ind, May 28, 2025, No. 2:24-cv-00326-JPH-MJD) [(involving six nonexistent cases across multiple filings; total sanction $6,000)]; Kruse v. Karlen, 692 SW3d 43, 50-54 (Mo Ct App 2024) (involving 22 fabricated case citations and multiple statutory misstatements by attorney; total sanction $10,000).
[9] See ABA Formal Opinion 512, Generative Artificial Intelligence Tools, July 29, 2024, ABA, https://www.americanbar.org/content/dam/aba/administrative/professional_responsibility/ethics-opinions/aba-formal-opinion-512.pdf; see also, ABA Issues First Ethics Guidance on a Lawyer’s Use of AI Tools, ABA, July 29, 2024, https://www.americanbar.org/news/abanews/aba-news-archives/2024/07/aba-issues-first-ethics-guidance-ai-tools/.
[10] See id.
[11] See Analysis of AI Use in Courts, March 14, 2024, https://rails.legal/resources/resource-ai-orders/.
