Introduction: The Dominance of Artificial Intelligence
Since its inception, artificial intelligence (“AI”) has rapidly embedded itself into the daily life of the everyday user and the legal practitioner. From drafting and document review, to research acceleration and analytics, AI generative tools are now common in the legal profession. The promise is efficiency and insight; the peril is error at machine speed and scale. For New York practitioners, the line between responsible usage and professional misconduct is now clearly demonstrated by the recent case of Landberg v. City of New York.[1] Indeed, the Landberg decision – issued by the Second Department – provides a defining moment: it imposes substantial sanctions for using AI-fabricated citations and misstatements, it condemns an attorney’s lack of candor when confronted by the court about improper use of and reliance on AI, and it admonishes opposing counsel for failing to alert the court to fake authorities.
The Dangers of AI Hallucinations
Since the introduction AI into the legal world, lawyers have been lectured on the many dangers of using artificial intelligence. These dangers include the fact that generative AI excels at confident synthesis, which leads users to trust the tool’s accuracy. However, this trust may be misplaced because AI hallucinates citations, quotes, and analyses. The overreliance on AI, paired with the failure to verify one’s work, has led to courts facing needless burdens, adversaries incurring avoidable costs, and clients being prejudiced. The professional hazard is severe, as even a few unverified cases and quotations can metastasize into multiple rule violations after briefs containing fabricated information are filed with the court and the case argued.
The Landberg decision now adds clarity to the use of AI: the submission of fabricated citations, fictitious quotes, and misstatements of holdings is sanctionable. Landberg puts that principle into sharp relief and sets new, severe benchmarks for penalties.
More specifically, in Landberg, the brief of the plaintiff-appellant was “prepared with the assistance of generative artificial intelligence” and included citations to nonexistent cases, and fictitious quotations attributed to the Court of Appeals that were contrary to actual law.[2] Beyond containing fabricated cases and quotations, the brief also misrepresented holdings of real cases.
Aside from discussing the extreme impropriety of appellant submitting hallucinated cases in its brief, the Landberg court addressed the candor of plaintiff’s appellate counsel.[3] That is, at oral argument, when the judicial panel raised concerns about the use of fictitious and erroneous citations and quotations in plaintiff’s brief, and offered appellant’s counsel an opportunity to explain the fictitious material, counsel indicated he was “not prepared to speak” about the cases at issue. The court even offered appellate counsel a 15-minute recess to locate sources to support the questionable citations. Counsel declined that offer too and, instead, stated that he personally wrote the brief and that the authorities came from Lexis, Westlaw, a book, or prior briefs. Subsequently, after oral argument, and in response to the court’s order to show cause whether sanctions should be imposed against appellant’s counsel and firm, appellate counsel acknowledged he should have said the cases at issue were hallucinated by AI.[4] The appellant’s attorney’s firm also submitted an affirmation to the court, acknowledging the severity of appellate counsel’s error, but also stating that the firm believed the attorney was genuinely remorseful.[5]
The Sanctions in Landberg
The court identified the recognized pitfalls of AI—nonexistent case citations, fabricated quotations from actual cases, and misstatements of law – and concluded that the filing of a brief containing such defects, coupled with counsel’s lack of candor, warranted sanctions under 22 NYCRR 130-1.1.[6] The Landberg court stated, “In determining an appropriate sanction, courts must consider that sanctions are both retributive, in that they punish past conduct, but are also goal oriented, in that they are useful in deterring future frivolous conduct not only by the particular parties, but also by the [b]ar at large.”[7]
The court sanctioned the individual attorney $8,000, emphasizing his lack of candor and submission of fabricated quotations that significantly misstated the law, along with other misstatements that misrepresented real holdings.
The court also sanctioned the law firm $2,500. Although crediting the firm’s statements that counsel’s conduct contravened firm policy, the court determined firm-level sanctions were appropriate for deterrence because the firm’s name appeared on the misrepresentative brief.
Landberg also highlighted a systemic safeguard that failed in this instance: the duty of opposing counsel – as officers of the court – to assist the court by reporting the suspicion of hallucinated cases and helping police the record. During oral argument and in its decision, the court noted that although opposing counsel filed respondents’ briefs, they failed to alert the court to the fabricated citations, quotations, and misstatements in the appellant’s submission.[8]
Landberg Takeaways and Practical Implications for Practitioners
Verification is non-delegable. AI can assist, but it cannot own, the authority in your papers. Landberg offers concrete lessons for drafting, supervision, training, and remediation.
Verify authority before filing. Landberg condemns reliance on AI outputs without human verification to check for, and guard against, core risks that might materialize in the appellant’s brief as a result of AI use: hallucinated case citations, fabricated quotations, and misstatements of holdings.
Supervision and firm policy. The court credited the law firm’s policy requiring personal review and confirmation of AI-generated citations and noted the firm’s post-incident ethics training and warnings, yet still imposed a firm sanction for deterrence. An attorney’s sign-off should include a verification certification for all authorities in court filings.
Oral argument readiness and candor. When confronted about fabricated authorities, plaintiff’s counsel declined a recess and instead offered shifting explanations rather than acknowledging AI involvement. The court later found that counsel was not being candid, and deemed this professional immaturity and disrespect for the system.[9]
Remediation when errors are discovered. If you learn of a defect, notify the court and adversaries immediately with a corrected brief or letter, explaining precisely what changed and why. Landberg shows that minimizing or compartmentalizing the problem compounds it.
Peer review. Have someone other than the drafter independently re-check every citation, quotation, and proposition before filing.
Tool usage. Use reputable, accountable platforms with citation retrieval features. However, even with such tools, read the primary source. Disable training on confidential inputs, and maintain logs of when, how, and for what tasks AI was used.
The Role of Opposing Counsel: Flagging Errors to Protect the System
As demonstrated in Landberg, the duty to detect fabricated cases, quotations, and citations does not rest solely on the attorney who misuses AI in writing a brief, but also upon opposing counsel. That is, although respondents’ attorneys in Landberg filed briefs in support of their own position, they did not alert the court to the fake citations, fabricated quotations, and misstatements contained in the appellant’s brief. During oral argument, the court chastised respondents’ attorneys for their conduct, and admonished them for allegedly reviewing appellant’s brief and failing to find, and report to the court, the glaring misstatements of law, and fictitious case citations and quotations presented in appellant’s brief.
Ultimately, although the lawyer who misuses AI bears the primary responsibility, the lawyer who fails to point out the misuse of AI now also carries responsibilities. This additional layer of review is the court’s latest attempt to adjust the standard of practice in the legal world regarding AI. Thus, when opposing counsel encounters suspicious authorities, said attorney should (i) verify immediately on a trusted database, (ii) notify opposing counsel promptly and request correction if misrepresentation or fake cases or quotations are found, (iii) alert the court by letter or motion, as appropriate, if the errors are not corrected, and identify the issue precisely and attach verified materials. This protects the record, preserves resources, and may factor into sanctions and fee-shifting analyses. It will also demonstrate to the court and your client that you are carefully reviewing opposing counsel’s papers.
Conclusion
The temptation is great: faster drafting, instant case authorities, and synthesized argumentation. Despite the great advances in AI, the price of unverified reliance is greater still. Landberg crystallizes New York’s posture on AI: fabricated citations and quotes, misstatements of holdings, and evasions when questioned by the court about suspected misuse of AI will draw severe sanctions against both attorney and firm. The path forward is equally clear: rigorous verification workflows, genuine supervision, transparent remediation, and a shared commitment by advocates on both sides, as officers of the court, to protect the integrity of the judicial process.
[1] Landberg v. City of New York, 2026 N.Y. Slip Op. 03935 (2d Dept. June 23, 2026).
[2] Id. at *1.
[3] Id. at *2.
[4] Id.
[5] Id. at *4.
[6] 22 NYCRR 130-1.1 (A court “in its discretion, may award to any party or attorney in any civil action or proceeding before the court, except where prohibited by law, costs in the form of reimbursement for actual expenses reasonably incurred and reasonable attorney’s fees, resulting from frivolous conduct”); see also Landberg at *4-*5.
[7] Id. at *4, quoting Deutsche Bank Nat’l Trust Co. v. LeTennier, 250 N.Y.S.3d 260, 268 (3d Dept. 2026) (internal quotations omitted).
[8] Landberg at *2.
[9] Id. at *6.