Flycatcher v. Affable Avenue (S.D.N.Y.): a lawyer's repeated AI-fabricated citations cost his client the case by default judgment, then cost him $38,877.59 in fees

The problem

Flycatcher Corp. Ltd. and Flycatcher Toys, Inc. sued Affable Avenue LLC and other defendants in the Southern District of New York on December 11, 2024, over trademark infringement and related claims [source]. Affable’s lawyer, Steven A. Feldman, filed a motion-to-dismiss brief in June 2025, and on June 26, 2025 counsel for co-defendant Top Experience, Joel MacMull, told the court it cited “at least thirteen cases Mr. Feldman cited that did not exist, and eight cases that did exist but did not contain the quotes Mr. Feldman attributed to them” ([source]). Those counts are MacMull’s, as recited by the court, not a tally the court made itself [source]. At the August 22, 2025 hearing Feldman corrected the judge’s “one-quarter” with his own figure, “Fourteen out of 60 cited cases,” and the court answered that “it is 14 fake cases too many” [source]. Ars Technica restates that as “three filings containing 14 errors out of 60 total citations,” which is looser than the opinion: there, the 14-of-60 figure belongs to the first brief alone ([source]).

What was built

Nothing was built. This is an honest-negative about what happens when AI output goes to a court unchecked. Feldman told the court he found cases through Google Scholar, Google and vLex and then cite-checked by running the brief through “either vLex or Paxton AI,” and under oath he agreed he had not cite-checked it himself: “The Court: You didn’t fully cite check the brief, before you submitted it to me. Mr. Feldman: Correct, I did not.” [source]. His response to the Order to Show Cause, which quoted Ray Bradbury and invoked “the ancient libraries of Ashurbanipal,” itself carried a quotation attributed to Mata v. Avianca that the court traced to a 2023 law-firm article, and for that filing he said the tool he used was NotebookLM [source]. His August 8, 2025 proposed reply brief then cited a Himmelstein case under reporter citations that, the court found, do not exist [source]. The court’s finding: “Mr. Feldman misused AI in three separate filings, which resulted in the submission of nonexistent cases and misattributed quotes” [source]. The court named the tools but blamed none of them, stating that “its problems with Affable’s submissions are not the use of AI per se” ([source]).

The outcome

On February 5, 2026, finding bad faith, Judge Failla imposed a terminal sanction: “the Court SANCTIONS Mr. Feldman by entering default judgment as to his client, Defendant Affable Avenue LLC,” struck docket entries 153 through 156, and deferred the assessment of damages until the claims against the remaining defendants are resolved [source]. The court explained why the client bears it: a party “cannot ‘avoid the consequences of [its lawyer’s] acts or omissions’,” and “If Mr. Feldman could not verify a citation, he should not have cited it” ([source]). Justia News and Ars Technica reported the default judgment independently [source] ([source]).

On October 9, 2026, the court priced the lawyer’s share. MacMull sought $43,086.59: “$42,090.00 for 73.2 hours of attorney time billed and $996.59 in Westlaw legal research fees,” at $575 an hour [source]. “[L]argely motivated by Mr. Feldman’s limited ability to pay,” the court cut the fee component by 10% and awarded MacMull’s firm, Mandelbaum Barrett PC, $38,877.59, payable by Feldman “within 30 days of this Opinion” [source]. The figures reconcile: $42,090.00 less 10% is $37,881.00, plus $996.59 is $38,877.59 ([source]). Two independent reports carry the award: USA Herald, a general-news site and the weakest source on this page, so it confirms the figure rather than establishes it [source], and a National Law Review analysis by a Greenberg Traurig attorney, a firm not involved in the case ([source]). The court also justified paying the lawyer who reported the errors: “courts depend on attorneys to bring to their attention other lawyers’ faulty citations” [source].

How this was verified Method: independent audit of the public record only. No party was contacted. Both signed opinions, the February 5, 2026 sanctions opinion (ECF 227, 33 pages) and the October 9, 2026 fee opinion (ECF 364, 18 pages), were read in full from CourtListener RECAP, and each PDF is byte-identical (sha1) to its Wayback capture. Every quoted figure was matched verbatim against the extracted text. The default judgment is corroborated by Reason/The Volokh Conspiracy, Justia News and Ars Technica, and the fee award by USA Herald and The National Law Review, each read from its own archived capture. The 13 + 8 count is opposing counsel’s and the 14-of-60 count is the lawyer’s own; the story labels both that way. Verification date: 2026-10-20.

Related case files

• Mata v. Avianca, the canonical $5,000 ChatGPT-citation sanction in the same court that Feldman tried to distinguish himself from
• Couvrette v. Wisnovsky, the other case where AI-fabricated citations ended the case, by dismissal with prejudice plus $15,500
• Lacey v. State Farm, where a special master also shifted the cost of AI-fabricated citations as fees, $31,100
• Deutsche Bank v. LeTennier, New York’s first appellate AI sanction, where counsel also kept filing fake cases after being caught