In a trademark action pending in the Southern District of New York, Judge Failla addressed defense counsel’s repeated filing of briefs containing fake, AI-generated case citations. There are two opinions worth discussing.
- The Sanctions Opinion — Feb. 5, 2026, 2026 WL 306683. Here, the court found bad-faith Rule 11 violations and imposed terminal sanctions (default judgment) on the defendant (Flycatcher Corp. v. Affable Ave. LLC, Case No. 24 CIV. 9429 (KPF), 2026 WL 306683 (S.D.N.Y. Feb. 5, 2026)).
- The Fee Opinion — Oct. 9, 2026, 2026 WL 3033675, issued this month, awarding the reporting attorney $38,877.59 in fees and costs. (Flycatcher Corp. Ltd. & Flycatcher Toys, Inc., v. Ozmos Co. LLC et al., Case No. 24 CIV. 9429 (KPF), 2026 WL 3033675 (S.D.N.Y. Oct. 9, 2026)).
Together, the two decisions illustrate the full arc of an AI-hallucination sanctions matter: detecting the hallucination, a bad-faith finding, a case-terminating sanction against the client, and monetary fee-shifting personally against the offending lawyer.
The Feb. 5, 2026, Sanctions Opinion
Counsel’s Use of Unverified AI-Generated Case Citations
Judge Failla found that counsel misused generative AI across three separate filings in support of his client’s motion to dismiss. In June 2025, he filed a brief “peppered with false citations.” After the court issued an order to show cause, he responded with a submission that appeared to have been AI-generated and that itself contained a false citation — prompting the court’s warning that he was “not excused from [the] professional obligation [of verifying his cases] by dint of using emerging technology.” He then filed a proposed reply brief containing yet another nonexistent case while the show-cause matter was pending. (Flycatcher Corp. v. Affable Ave. LLC, Case No. 24 CIV. 9429 (KPF), 2026 WL 306683 (S.D.N.Y. Feb. 5, 2026)).
At an Aug. 22, 2025 sanctions conference, counsel — placed under oath — admitted relying on AI but gave discursive, often non-responsive answers. The court found that 14 of the 60 cases cited in his motion-to-dismiss brief were nonexistent AI hallucinations, with misattributed quotes in others. (Flycatcher Corp. v. Affable Ave. LLC, Case No. 24 CIV. 9429 (KPF), 2026 WL 306683 (S.D.N.Y. Feb. 5, 2026)). By his own account, counsel located cases via an internet search and then ran his brief through three rounds of AI review and never personally cite-checked the brief. (Flycatcher Corp. v. Affable Ave. LLC, Case No. 24 CIV. 9429 (KPF), 2026 WL 306683 (S.D.N.Y. Feb. 5, 2026)).
Rule 11 and the Bad-Faith Standard for AI Citation Sanctions
The court drew its sanctions authority from two sources — Rule 11 and the court’s inherent powers. It emphasized that to impose sanctions under either source sua sponte, it must find subjective bad faith and a heightened mens rea standard. (Flycatcher Corp. v. Affable Ave. LLC, Case No. 24 CIV. 9429 (KPF), 2026 WL 306683 (S.D.N.Y. Feb. 5, 2026)).
Critically, the court did not condemn AI use as such. It stated it “does not oppose the use of AI to assist in legal research and writing,” but “must take a stand where …counsel repeatedly files submissions with false citations because counsel refuses to verify those submissions.” Id.
Why the Court Found Bad Faith and Imposed Default Judgment
The court found that counsel’s AI misuse, “exacerbated by his insouciant approach to cite-checking, was done in bad faith” — he knew, or consciously avoided learning, that his cited cases were not warranted by existing law, and knew or was aware of the high probability that using AI as he did would generate faulty citations. Even fully crediting his explanation, the court held, he “would still have violated Rule 11 by submitting cases without reading them,” invoking the Second Circuit’s rule in Park v. Kim that attorneys must read and confirm the authorities on which they rely. 91 F.4th 610 (2d Cir. 2024))
Weighing the five factors governing terminal sanctions — intentional bad faith, prejudice, pattern, whether corrected, and likelihood of recurrence — the court found the dispositive consideration to be counsel’s continuous pattern: he kept submitting fake citations and refusing to verify them despite repeated warnings and despite his admitted ability to access legal research tools through a bar library. Accordingly, the court struck defendant’s submissions and entered default judgment against it, and granted plaintiff’s counsel leave to seek fees under Rule 11(b) and 28 U.S.C. § 1927, having found bad faith and that the behavior “multiplie[d] the proceedings …unreasonably and vexatiously.” (Flycatcher Corp. v. Affable Ave. LLC, Case No. 24 CIV. 9429 (KPF), 2026 WL 306683 (S.D.N.Y. Feb. 5, 2026)).
The Oct. 9, 2026, Fee Opinion
With entitlement to sanctions established, on Oct. 9 the court rendered its fees determination. While the attorney who had uncovered and reported defense counsel’s fake citations sought $43,086.59 in fees and costs after being given “the runaround” in efforts to resolve the fee informally, the court applied the lodestar method and found the $575 hourly rate and the hours (including time investigating the misconduct and preparing the fee petition) generally reasonable. It rejected defense counsel’s argument that the work was unnecessary, reasoning that as AI citation problems proliferate, courts depend on attorneys to bring opposing counsel’s faulty citations to their attention and crediting the investigation as a service that saved the court substantial time. (Flycatcher Corp. Ltd. & Flycatcher Toys, Inc., v. Ozmos Co. LLC et al., Case No. 24 CIV. 9429 (KPF), 2026 WL 3033675 (S.D.N.Y. Oct. 9, 2026)).
The court then imposed a 10% across-the-board reduction to the fees component — motivated largely by defense counsel’s limited ability to pay as a solo practitioner, under the principle that a court may temper a sanction by the offending attorney’s ability to pay — yielding a total award of $38,877.59, payable within 30 days.
Takeaways
As with most decisions involving AI, the decisions signal that using AI is permitted; but failing to verify its output is sanctionable. Judge Failla was explicit that AI-assisted research and writing is acceptable. The violation was the non-delegable duty of verification — the same principle the Second Circuit announced in Park v. Kim.
The decisions also underscore that bad faith may rest on conscious avoidance. The court did not need to find that defense counsel knew his citations were fake. It was enough that he was aware of a high probability that his AI-driven process would generate fabrications and proceeded anyway — a conscious-avoidance theory of subjective bad faith. Moreover, the pattern and persistence of the offending behavior drove the severity. Indeed, what moved the court from lesser measures to the “extraordinary” step of a client-terminating default was repetition of improperly using AI in the face of explicit warnings — three filings, each with fabricated authority, including one filed while the sanctions show-cause order was pending.
Notably, these decisions show that courts are rewarding opposing counsel who catch hallucinations, effectivelydeputizing one’s adversary. By compensating plaintiff’s counsel’s investigation — and framing it as a service to the court — Judge Failla may have created a financial incentive for opposing counsel to police AI-generated filings and shifted that cost onto the offending lawyer personally.
Recent Comments