“Incredulity” That “Competent Law Firm Representing One of the Largest Financial Institution[s] in the World” Filed Brief with AI Hallucinations

From Douglas v. Deutsche Bank Nat’l Trust Co., decided yesterday by D.C. Court of Appeals Judges Catharine Easterly, Vijay Shanker, and Stephen Glickman:

What began as a routine appeal has become a cautionary tale about the misuse of artificial intelligence (AI) and its consequences for attorneys, their clients, and the courts…. While reviewing appellee’s [i.e., the Bank’s] brief, the court discovered that it contained multiple citations to cases that the court was unable to locate or confirm as legitimate. On June 22, we issued an order requiring appellee to show cause “why the court should not strike its brief for citing nonexistent cases that are possibly the product of artificial intelligence (AI) hallucinations.”

The next day, one of appellee’s attorneys at the firm, Loishirl W. Hall, filed a response in her own capacity. Ms. Hall confirmed that four of the brief’s cited authorities did not exist. She acknowledged that these citations were “not legitimate legal authority” and “should not have appeared in a brief filed with this Court.” {Ms. Hall has informed the court that she no longer works at the firm and thus no longer represents appellee, although appellee has not moved for her withdrawal from the case.}

Ms. Hall explained that she had “used Google’s generative artificial intelligence search tool to assist in locating case authority” and did not verify the existence or accuracy of those citations before filing the brief. She recognized that she had a duty to verify the accuracy of every authority submitted to this court and apologized for her failure to do so. As part of this mea culpa, Ms. Hall assured the court that she had taken the remedial measure of verifying that the remaining citations in the brief were legitimate authority and pledged that, going forward (presumably in her representation of other clients), she would not rely on any AI tool “for legal research without confirming each authority” or file anything that includes citations she has not independently verified. Ms. Hall represented that she “deeply regrets this error and the burden it has placed on the [c]ourt and all parties.”

The firm subsequently filed a response for appellee. The firm stated that it was “unaware” of Ms. Hall’s actions when the brief was filed. Although two other attorneys—Jianna Jaques Santos and Michael T. Cantrell—were listed on the brief with Ms. Hall, no information was provided about their review of appellee’s brief before filing. Instead, the firm represented that it “only became fully aware of the situation” (emphasis added) once the court issued its show-cause order and Ms. Hall responded. The implication—that the firm did nothing after the court issued its order to become “fully aware” on its own that it had included fake case citations in its brief—is surprising, to say the least. The firm then shifted much of the blame to Ms. Hall, who had stated in her filing that she had never filed an appellate brief before this one.

The firm represented that it prohibits employees from using AI in the “drafting of any legal correspondence or documents” and that failing to verify citations “obtained in any manner, which includes the use of artificial intelligence, is a violation of firm policy” of which all employees are made aware during initial and annual employment trainings. The firm did not attach the policy for this court’s review. Highlighting that “these actions were taken by a former employee” of the firm, and without detailing what actions the firm had taken to supervise or review Ms. Hall’s work, the firm acknowledged only that “further review should have been undertaken.” The firm then assured the court that it was “genuinely remorseful for this failure” and that it intends to take appropriate measures to ensure that this “type of situation” never occurs again.

In replying to Ms. Hall’s response, Mr. Douglas asked the court to, among other things, deny Deutsche Bank an opportunity to file a corrected brief. He registered incredulity at the fact that “a competent law firm representing one of the largest financial institution[s] in the world could make” such a mistake. Mr. Douglas did not file a separate reply to appellee’s response to the show-cause order.

As we discuss below, we echo Mr. Douglas’s incredulity….

You can read the full opinion for more details, but here’s the bottom line as to sanctions:

To be sure, some of the cases cited in Deutsche Bank’s response brief are genuine and some of the propositions of law are sound. But we “reject any notion that because some authority exists to support a legal proposition it should negate the harm caused by hallucinated cases. Put bluntly—absolutely not; period; end of story; all stop.” We caution that citing to even a single fake case can be sanctionable because “no brief, pleading, motion, or any other paper filed in any court should contain any citations—whether provided by generative AI or any other source—that” an attorney has not personally “read and verified.” Competent legal representation requires the investment of significant time and effort; if an attorney deems their client’s claims as “not worth the effort, [they] should either tell [their] clients that they fall short or advise them to hire another lawyer—not file obviously insufficient pleadings and briefs generated by AI tools.”

It is axiomatic that our rules exist to ensure the orderly, prompt, and just disposition of appeals. These rules exist not for the convenience of the court but for the benefit of litigants and counsel who have a right to rely on them and must comply with them. We cannot condone disregard of court rules, whether willful or negligent. As such, we deem it appropriate to strike appellee’s brief. {The firm suggested that the court could accept the appellee’s brief as filed but with the erroneous citations redacted. We reject this suggestion, as it would constitute effectively no sanction at all.}

We also refer this matter to the Office of Disciplinary Counsel for any investigation it deems appropriate.

Judge Glickman concurred to discuss the scope of D.C. courts’ sanctioning authority, but also added this (as well as another matter that I note in a separate post):

Bad as AI-fabricated citations are, we should recognize that they are only the tip of a perilous iceberg. That iceberg is the use of artificial intelligence not only to research the law but also to draft a brief. This temptation is the more serious evil, and not just because of hallucinated citations.

To begin with, the fallibility of AI as a tool for lawyers is by no means limited to invalid citations. Notably, as the Ninth Circuit explained in Malkeet Lnu v. Blanche (9th Cir. 2026), generative AI “is prone to make, at least in its current stage of advancement,” two types of mistakes—not only fabrications, but also what the court described as “inaccuracies”:

Fabrications are instances in which the generative AI tool provides cases or quotations that do not exist at all. Inaccuracies are more subtle. The generative AI tool might cite to real authorities but provide an answer that is legally or factually inaccurate or not supported by the citation.

Fabrications are the most notorious hallucinations, but inaccuracies may prove more dangerous to our profession in the long run. Inaccuracies are more likely to go unnoticed by attorneys and judges because they are not always susceptible to facial checks. Rather, identifying these misunderstandings often requires close analysis of cited sources. With close analysis, some inaccuracies might be clear—for instance, claiming that a case expressly stands for a proposition about a certain topic when the case does not discuss that topic at all. Others may be difficult to distinguish from poor legal reasoning.

And inaccuracies are common, even in newer generation models that produce fewer fabrications. Including inaccuracies, legal-specific generative AI tools from Westlaw and Lexis hallucinated 17% and 33% of answers, respectively, to a representative set of queries run in 2024. The most common error modes of the latest generation tools include misunderstanding holdings, failing to distinguish between legal actors (e.g., presenting a rejected party argument as the holding of the court), and failing to respect the hierarchy of authorities. In other words, the sort of errors that we might expect a first-semester law student to make, but certainly not licensed attorneys appearing before this court.

Even worse, perhaps, is the fact that relying on AI to do the “thinking” necessary to draft and refine a brief—even what is intended to be only a first draft—is an abdication of the lawyer’s duty and a breach of the client’s (and the public’s) trust. Lawyers are entrusted with the responsibility to use their own unique human experience, insight, and creativity in employing their skills, learning, and judgment to write, advocate and persuade; and, in doing such writing, to choose among options and alternatives; to select and comprehend precedents and authorities; to make moral and ethical decisions; to evaluate which claims and contentions to pursue and which not; and so forth. AI cannot duplicate that multifarious exercise of human judgment and ingenuity in the “thinking” enterprise.

We do not know in this case whether Ms. Hall relied on Google not only to find case citations, but to draft her brief. I hope she did not do so, and that lawyers will refrain from such conduct.