The Volokh Conspiracy
Mostly law professors | Sometimes contrarian | Often libertarian | Always independent
No Sanctions in Michael Cohen Hallucinated Citations Matter
From today's decision in U.S. v. Cohen by Judge Jesse Furman (S.D.N.Y.) (see also N.Y. Times [Benjamin Weiser]):
In support of his motion [for early termination of supervised release], Schwartz [Cohen's lawyer] cited and described three "examples" of decisions granting early termination of supervised release that were allegedly affirmed by the Second Circuit. See id. at 2-3 (citing United States v. Figueroa-Florez, 64 F.4th 223 (2d Cir. 2022); United States v. Ortiz (No. 21-3391), 2022 WL 4424741 (2d Cir. Oct. 11, 2022); and United States v. Amato, 2022 WL 1669877 (2d Cir. May 10, 2022)). There was only one problem: The cases do not exist. Although the Government failed to point that fact out in its opposition to Cohen's motion, E. Danya Perry—who entered a notice of appearance on Cohen's behalf following the Government's submission—disclosed in a reply that she had been "unable to verify" the citations in Schwartz's filing….
Schwartz (aided by his own counsel) and Cohen (aided by Perry)[,] … [w]ith one exception discussed below, … tell the same basic story. In early November 2023, Schwartz sent a draft of what would become the November 29, 2023 motion to Cohen. Cohen asked Perry (who had not yet entered an appearance in this case) to provide feedback on the draft, which she did. One comment, which Cohen passed along to Schwartz, was that the motion should cite a few cases granting early termination. Schwartz adopted what he understood to be Perry's suggestions and sent subsequent drafts back to Cohen.
On November 25, 2023, Cohen then sent three emails to Schwartz with the cases in question and summaries of the cases. Cohen had obtained the cases and summaries from Google Bard, which he "did not realize … was a generative text service that, like Chat-GPT, could show citations and descriptions that looked real but actually were not. Instead, [he had] understood it to be a super-charged search engine …." According to Cohen, he did not "have access to Westlaw or other standard resources for confirming the details of cases" and "trusted Mr. Schwartz and his team to vet [his] suggested additions before incorporating them" into what became the motion.
That trust proved unfounded.
Although Cohen's November 25, 2023 emails made no reference to Perry, Schwartz "believed" that the cases "had been found by Ms. Perry" given the earlier back and forth. "[B]ecause of Ms. Perry's reputation" as "a renowned and skilled trial lawyer," Schwartz "did not independently review the cases." He notes that he "never contemplated that the cases cited were 'non-existent,'" that their inclusion in his motion was an "honest" and "unfortunate mistake[],"and that he "had no intention to deceive the Court."
The first Schwartz learned of the problem was when the Court issued its Order to Show Cause on December 12, 2023; he then spoke to Perry's colleague, who informed him that Cohen, not Perry, had been the source of the cases. Schwartz professes that he "would have researched" the cases had he "believed that Mr. Cohen" was their source. He acknowledges that he "bear[s] the responsibility for" the "submission" and that "the inaccuracies" in it were "completely unacceptable," and he "sincerely apologize[s] … for not checking [the] cases personally before submitting them."
As noted, Schwartz's and Cohen's recollections differ in one respect.Schwartz recalls that Cohen had, before sending him the cases on November 25, 2023, "communicated to [him] that cases would be provided by Ms. Perry," which was one reason he believed the cases came from Perry.Cohen denies that he said anything of the sort to Schwartz, although he does not dispute Schwartz's claim that Schwartz mistakenly believed in good faith that the cases had come from Perry.
In the Court's view, this discrepancy is ultimately inconsequential. At most, Cohen told Schwartz on some unspecified date that Perry would later provide "cases"; Schwartz does not suggest, and the record does not show, that Cohen identified Perry as the source of the problematic cases Cohen provided on November 25, 2023. Moreover, even if Cohen did not say anything to Schwartz to suggest that Perry would provide cases, Perry's comments on the initial draft that Cohen forwarded to Schwartz provided a good faith basis for Schwartz's belief that Perry was the source. All of that is to say, the Court credits Schwartz's testimony that he genuinely, but mistakenly, believed that the cases had come from Perry; that he did not independently review the cases based on that belief; that he would have researched the cases had he known that Cohen was the source; and that he did not intend to deceive the Court….
Judge Furman concluded that sanctions should only be imposed for knowing misbehavior, and not carelessness, and that there wasn't enough evidence here of such knowing misbehavior:
[T]he Court is compelled to conclude that sanctions may not be imposed on Schwartz. His citation to non-existent cases is embarrassing and certainly negligent, perhaps even grossly negligent. But the Court cannot find that it was done in bad faith. Given Perry's comments on the initial draft (as conveyed by Cohen), Schwartz understandably believed that the cases had come from her. As Schwartz himself concedes, it was plainly his responsibility to review the citations before putting them in a submission to the Court. But the Court credits his explanation for his failure to do so: that he had confidence in the accuracy of the cases given Perry's reputation and that he never contemplated that the cases were non-existent.
The Court also credits Schwartz's representations—that inclusion of the cases in his motion was an "honest" and "unfortunate mistake[]"; that he "had no intention to deceive the Court"; and that he would have withdrawn the citations immediately if given the opportunity—and does not doubt the genuineness of his apologies and acceptance of responsibility. Perry, in responding to Schwartz's submission, asserts that "even a quick read" of the citations "should have raised an eyebrow." That may be so, but, at most, it supports a finding of extreme carelessness, not intentional bad faith. And it is noteworthy that the Government's lawyers—who presumably engaged in more than a "quick read" of Schwartz's motion and, thus, saw the citations at issue before filing their opposition—did not "notice[] something awry" themselves.
So too, there is no basis to impose sanctions on Cohen himself. The Court's Order to Show Cause was limited to Schwartz and did not alert Cohen to the possibility of sanctions. But even if the Court had put Cohen on notice, sanctions would not be warranted. Cohen is a party to this case and, as a disbarred attorney, is not an officer of the Court like Schwartz. He was entitled to rely on his counsel and to trust his counsel's professional judgment—as he did throughout this case.
Given the amount of press and attention that Google Bard and other generative artificial intelligence tools have received, it is surprising that Cohen believed it to be a "super-charged search engine" rather than a "generative text service." But the Court has no basis to question Cohen's representation that he believed the cases to be real. Indeed, it would have been downright irrational for him to provide fake cases for Schwartz to include in the motion knowing they were fake—given the probability that Schwartz would discover the problem himself and not include the cases in the motion (as he should have) or, failing that, that the issue would be discovered by the Government or Court, with potentially serious adverse consequences for Cohen himself.
In sum, as embarrassing as this unfortunate episode was for Schwartz, if not Cohen, the record does not support the imposition of sanctions in this case.
Note that sanctions sometimes do get imposed for merely careless behavior. My sense is here the judge was especially influenced by this detail:
All of that is to say, the Court credits Schwartz's testimony that he genuinely, but mistakenly, believed that the cases had come from Perry; that he did not independently review the cases based on that belief; that he would have researched the cases had he known that Cohen was the source; and that he did not intend to deceive the Court.
If in a future case a lawyer instead just relies directly on ChatGPT or Google Gemini, without cite-checking—as opposed to relying on information that he thought came from another, reputable lawyer—the result might well be different (as it was in Mata v. Avianca, the first hallucitation court case to hit the news).
To get the Volokh Conspiracy Daily e-mail, please sign up here.
Show Comments (26)