D.C. Court of Appeals strikes Deutsche Bank brief over AI-hallucinated citations

What the court ordered

On September 3, 2026 the District of Columbia Court of Appeals filed a published order in Douglas v. Deutsche Bank National Trust Co., No. 24-CV-1099, striking the appellee’s brief. The appeal arises from a Superior Court judgment on the pleadings for Deutsche Bank National Trust Company, as trustee, in a judicial foreclosure under D.C. Code § 42-816. Deutsche Bank was represented by McCabe, Weisberg, & Conway, LLC; appellant Barry Douglas was self-represented.

While reviewing the brief after submission without argument on March 11, 2026, the panel found citations it could not locate. On June 22, 2026 the court ordered Deutsche Bank to show cause why the brief should not be struck for citing nonexistent cases possibly produced by AI hallucinations. Attorney Loishirl W. Hall, then at the firm, responded the next day in her own capacity and confirmed that four cited authorities did not exist. She said she had used Google’s generative artificial intelligence search tool to locate case authority and had not verified those citations before filing. The order lists the fake case names as Abadie v. District of Columbia, Cason v. Nat’l Consumer Co-op Bank, Osborne v. District of Columbia, and Woods v. United States, without reprinting the full false citations.

The firm later told the court it was unaware of Hall’s actions when the brief was filed, that two other attorneys were listed on the brief with Hall, and that firm policy prohibits using AI to draft legal correspondence or documents. The firm said it became “fully aware” only after the show-cause order and Hall’s response. Hall informed the court she no longer works at the firm.

Who it binds and what follows

The per curiam opinion (Associate Judges Easterly and Shanker, with Senior Judge Glickman) ordered the brief stricken and referred the matter to the Office of Disciplinary Counsel for any investigation it deems appropriate. The panel rejected the firm’s suggestion that the brief be accepted with erroneous citations redacted, calling that effectively no sanction. The court stressed that some genuine authorities and sound propositions remained in the brief, but that surviving real citations do not cure hallucinated ones. It cautioned that citing even a single fake case can be sanctionable because filings must contain only citations an attorney has personally read and verified.

The opinion states that every firm attorney who signed the brief bears some responsibility, and that attorneys can no longer credibly claim ignorance of generative AI’s propensity to invent authority. It also says the court does not seek to discourage AI use and that attorneys who eschew at least an understanding of AI do so at their peril.

Disagreement and the sanctions gap

Reporting on September 4, 2026 treated the order as a sua sponte appellate sanctions response and highlighted a rules gap. Senior Judge Glickman’s concurrence, which joins the result, argues that D.C. appellate rules and case law give the court narrower direct sanctioning tools than federal courts often invoke under Fed. R. App. P. 38 and 46 or inherent authority after a bad-faith finding. He characterizes striking the brief under D.C. App. R. 28 as appropriate but comparatively weak, and notes that on the existing record the conduct appears negligent rather than bad faith. He asks the full court and Rules Committee to consider clarifying sanctions authority, and he separately discusses how far each named brief signatory must personally cite-check every authority.

ABA Journal coverage attributed to Reuters that Hall admitted four citations were hallucinated by Google’s generative AI tool. The published order matches that admission and the Google-tool explanation. Law360’s September 4 report dated the panel’s order to Thursday (September 3) and framed the decision as flagging a sanctions-authority gap; the concurrence is the primary source for that gap analysis. Reason’s Volokh Conspiracy excerpted the incredulity language and the panel’s rejection of the idea that some real citations cancel hallucinated ones.

What remains unknown

The order does not decide whether any attorney violated the Rules of Professional Conduct; that question is left for disciplinary counsel. It does not resolve the foreclosure merits. It does not publish the full text of the firm’s AI policy. It does not state whether Hall used generative AI only for research or also to draft the brief. No financial penalty, suspension, or firm fine appears in the published order beyond striking the brief and the disciplinary referral.