Part of the AI Regulation News hub.
The District of Columbia Court of Appeals struck a bank's appellate brief and referred the matter to disciplinary counsel after the firm's attorney confirmed that four of the cited authorities did not exist
The remedy is the story. No fine, because on the concurring judge's reading the court may not have one available. What the court did instead was strike the brief, refuse to let the firm simply delete the bad citations, and say that every firm attorney whose name was on it bears some responsibility.
Bottom line: The court ordered that the appellee's brief be stricken and referred the matter to the Office of Disciplinary Counsel for any investigation it deems appropriate. It imposed no monetary sanction. The firm asked to keep the brief with the erroneous citations redacted; the court refused, stating that this would constitute effectively no sanction at all.
Who this affects: Appellate litigators in the District of Columbia, supervising partners and general counsel at firms that file there, and legal departments at institutional clients whose outside counsel use generative AI for research. The sanctioned side here was counsel for a national bank trustee, not a self-represented litigant.
Issue date: Published order filed 3 September 2026. The show-cause order that produced it issued on 22 June 2026.
What changed: No new rule was adopted. What is new is the disposition: a striking of the brief plus a disciplinary referral, and a published statement by the concurring judge that the court's own Rules appear not to give it the sanctioning powers the federal courts of appeals have been using.
Analysis: Read the concurrence. The brief was struck for violating D.C. Appellate Rule 28, which requires a party's contentions and reasons for them with citations to the authorities relied on. What Judge Glickman sets out, in a published document, is why nothing beyond that was available: why D.C. Appellate Rule 38 reaches frivolous appeals rather than briefs, and why Rule 46 is inapplicable to fabricated citations. He says the full court may wish to address the situation. He then does something more unusual: he describes the firm-side procedures he thinks would actually work, and says what is not required.
Primary sources: Published order, District of Columbia Court of Appeals, No. 24-CV-1099 · Court of Appeals opinions index, which lists No. 24-CV-1099 as a Published Order
- Case
- Barry Douglas v. Deutsche Bank National Trust Company, as Trustee for Soundview Home Loan Trust 2006-OPT3
- Number
- No. 24-CV-1099, on appeal from Superior Court No. 2023-CAB-005422
- Court and panel
- District of Columbia Court of Appeals. Easterly and Shanker, Associate Judges, and Glickman, Senior Judge. Per curiam, with a concurrence by Judge Glickman.
- Jurisdiction
- District of Columbia, United States
- Bindingness
- A published order of the District of Columbia's highest local court, disposing of a sanctions question in a live appeal. The court's own opinions index also lists No. 24-CV-1099 as a Published Order, so the styling is the court's, not this desk's inference. What this desk did not verify is how D.C. practice treats the precedential weight of a published order as against a published opinion, and it makes no claim about it.
- What was found
- Appellee's brief cited multiple cases the court could not locate. In response to the show-cause order, attorney Loishirl W. Hall confirmed that four of the brief's cited authorities did not exist, said she had used Google's generative artificial intelligence search tool to assist in locating case authority, and did not verify the citations before filing.
- Disposition
- Brief stricken. Matter referred to the Office of Disciplinary Counsel for any investigation it deems appropriate. Redaction of the erroneous citations rejected. No monetary sanction. The issue of the court's sanctions authority referred to the court's Rules Committee.
- Editorial Note
- Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.
- Primary source
- https://www.dccourts.gov/sites/default/files/2026-09/Douglas%20v.%20Deutsche%20Bank%20Nat%27l%20Trt%20Co.%2024-CV-1099%20ORDER.pdf
What the court actually ordered
The underlying appeal was a judicial foreclosure action under D.C. Code section 42-816. Deutsche Bank, the appellee, was represented by the law firm McCabe, Weisberg, and Conway, LLC. Mr. Douglas was self-represented. The appeal had been submitted for decision without argument on 11 March 2026.
While reviewing the appellee's brief, the court discovered that it contained multiple citations to cases that the court was unable to locate or confirm as legitimate. On 22 June it issued a show-cause order asking why it should not strike the brief for citing nonexistent cases that are possibly the product of artificial intelligence (AI) hallucinations.
One of the firm's attorneys responded in her own capacity the next day. She confirmed that four of the brief's cited authorities did not exist, acknowledged that these citations were not legitimate legal authority and should not have appeared in a brief filed with this Court, and explained that she had used Google's generative artificial intelligence search tool to assist in locating case authority without verifying the results.
The order is short on the operative point. It reads: ORDERED that appellee's brief is stricken. The court added that it also refers this matter to the Office of Disciplinary Counsel for any investigation it deems appropriate. Note the wording: the referral is of the matter, and the investigation is whatever that office decides is appropriate. Nothing in the order determines that anyone will be disciplined.
The redaction request, and why the refusal matters
This is the part with the clearest practical consequence, and it sits in a footnote.
The firm suggested that the court could accept the appellee's brief as filed but with the erroneous citations redacted. The court rejected the suggestion, stating that it would constitute effectively no sanction at all.
That closes a repair route that a firm in this position would obviously reach for. If the only cost of filing fabricated authority were deleting it once caught, the sanction would be the cost of being caught, which is close to zero. The court said so in one sentence and moved on.
The court also placed the harm on the record in its own terms. A brief containing fake citations, it wrote, masquerades as useful advocacy, but it misdirects the court, burns time and resources, and sows confusion while the court gets to the bottom of the matter. By then the damage is done: the court is left without real guidance from one side of the adversarial equation and the client is left without a credible advocate.
It is worth noticing who absorbed that cost here. The party whose brief was struck is a national bank trustee. The party left facing a court with one side's brief gone is a self-represented homeowner in a foreclosure appeal.
Every attorney whose name was on the brief
The order states, per curiam and in the court's own voice, that while the court acknowledges the submitting attorney's attempt to take ownership of her error, every firm attorney who signed the brief bears some responsibility. It quotes the Ninth Circuit in Malkeet Lnu v. Blanche for the proposition that a competent and diligent attorney cannot decline to vet citations, in a brief he signs, for substantive validity, meaning to read the cited authorities and ensure that they are on point. It then adds, behind a Cf. signal, a comparison to Superior Court Civil Rule 11(c)(1), under which, absent exceptional circumstances, a law firm must be held jointly responsible for a violation committed by its partner, associate, or employee.
Two things about that passage need care, because the difference is where a reader gets misled.
First, the comparison to Superior Court Civil Rule 11(c)(1) is introduced with a Cf. signal, and the concurrence confirms that this Rule is not a Rule of the appellate court. The panel did not apply a firm-liability rule; it pointed at one that governs the trial court.
Second, the court's statement is a general principle, and the concurring judge says explicitly that he agrees with it subject to a clarification as to what it must entail. The scope of that responsibility is not settled by this order. Footnote 4 refers the whole question of the court's sanctions authority to the court's Rules Committee for analysis and potential clarification, describing the position as unclear.
Anyone reporting this as the D.C. Court of Appeals holding law firms liable for their attorneys' hallucinated citations has read past the footnote.
The concurrence: a court saying out loud what it cannot do
Judge Glickman writes separately, and the concurrence is longer and more consequential than the order it accompanies.
He records that this is the first reported case in which this court has received a brief containing fabricated citations, and that the court has not hitherto addressed the scope of its authority to sanction attorneys for such submissions. That statement is his, in a concurring opinion, and it is scoped to reported cases in this court.
He then calls the sanction the court did impose an appropriate but comparatively weak response that unfortunately penalizes the innocent client for its counsel's offense, and works through why nothing stronger was available. Superior Court Civil Rule 11 authorises the Superior Court to sanction attorneys, but it is not a Rule of the appellate court. D.C. Appellate Rule 38 reaches frivolous appeals, petitions and motions, but not briefs, and the court's own test for frivolousness is stringent: wholly lacking in substance, such that a reasonable attorney would conclude that it is not based upon even a faint hope of success on the legal merits. On that test, he writes, Rule 38 would not support a sanction for filing a brief that contains fake citations but is not totally baseless. D.C. Appellate Rule 46, he says, is totally inapplicable to such fabricated citations.
The comparison he draws is with the federal appellate courts, which have been reaching for Federal Rules of Appellate Procedure 38 and 46 and their inherent authority. He cites the Seventh Circuit's survey in Perez-Castillo v. Blanche, the Sixth Circuit's decision in Whiting v. City of Athens, and the Fifth Circuit in Fletcher v. Experian Information Solutions, where Rule 46(c) discipline was said to include monetary sanctions.
On inherent authority he applies Chambers v. NASCO, which requires a finding of bad faith. On this record, he writes, it does not now appear that either the attorney or the firm acted in bad faith, vexatiously, wantonly, or for oppressive reasons; the conduct appears to have been negligent or grossly negligent rather than intentional, knowing or reckless. He notes that recklessness may be a close question given the publicity about AI in legal work and the firm's claim to have instructed her on its policy against reliance on AI, and adds that the firm has not shared its policy with the court, nor explained how it was communicated to its lawyers. Even so, he writes, the factual record necessary to support a finding of recklessness has not been made.
His conclusion is stated as a limit, not a certainty: beyond the admonishment and the striking of the brief, our Rules and case law appear to preclude us from sanctioning the attorneys. The full court, he suggests, may wish to address this situation, and the court may wish to consider adopting a Rule comparable to Superior Court Civil Rule 11 for appellate filings.
The part a managing partner should read
Having agreed that every signatory bears some responsibility, Judge Glickman spends several pages on what that cannot mean, and the result is the most operational passage in the document.
He starts with vocabulary. Appellate briefs commonly list several attorneys; ordinarily only one of them actually signs. For his purposes all the named attorneys are deemed to be signatories who stand behind the brief. That does not mean they all carry identical responsibility for it.
Requiring every signatory to personally cite check every brief would, he writes, be impractical, unduly burdensome, and unnecessarily duplicative given the availability of reasonable alternative procedures. His two illustrations are concrete. Where a firm assigns sections of a complex brief to different associates, each associate is responsible for the citations in his own assigned section, and cite checking is in general a task that the principal drafter or drafters must perform; that hardly means each associate must double check a section drafted by another. And a United States Attorney's Office brief typically names trial counsel, the Chief of the Criminal Appellate Section and the United States Attorney herself, none of whom can realistically cite check every brief that carries their name.
What he proposes instead is a short list. Supervisors can, and in his view should, ban outright the practice of asking AI to draft a brief from scratch. Firms and law departments may train and retrain attorneys in the proper use of AI, adopt clear policies with provision for discipline, and require attorneys to confirm that they have complied. Where citations and case descriptions are to be double checked before filing, that vetting may be done cost-effectively by a trained paralegal. Triple and quadruple checking by other attorneys, whether signatories or not, is uncalled for and would be unduly burdensome and costly.
He also names the risk he considers larger than fake citations. Using AI to draft the brief, rather than to find the law, is in his words the more serious evil, and he adopts the Ninth Circuit's distinction between fabrications, where the tool supplies cases or quotations that do not exist at all, and inaccuracies, where it cites real authorities but gives an answer that is legally wrong. Only the first kind is caught by checking whether a case exists.
What we did not verify
What was opened: the court's published order, retrieved directly from dccourts.gov, HTTP 200, 289,217 bytes, with a native text layer. Every quotation in this piece was character-matched against the extracted text of that file. No mirror was used and no OCR step was involved.
One retrieval note worth recording, because it will mislead anyone who checks the link with a script. dccourts.gov refuses clients that do not send a browser user-agent: the same URL returns a 179-byte HTTP 403 gateway page instead of the document. That page is a shell, not a finding, and the proof is that a deliberately nonexistent path on the same host returns a byte-identical 403 with the same SHA-256. A retrieval failure against that host therefore says nothing about whether the order exists. With an ordinary browser user-agent the URL returns the 289,217-byte PDF, and the copy this piece was written from is byte-identical to a fresh retrieval. A reader clicking the link in a browser gets the document; the court's opinions index is listed above as the durable alternative.
What was not opened: the appellee's brief, the show-cause response filed on 23 June 2026, the firm's internal AI policy, which the concurrence records was not shared with the court, and the underlying Superior Court record. The cases the order cites, including Perez-Castillo v. Blanche, Whiting v. City of Athens, Fletcher v. Experian and Malkeet Lnu v. Blanche, were not independently retrieved for this piece; they are reported here as the order cites them.
What this desk refuses to claim. That the order is precedential: the document establishes only that it is styled a published order, and D.C. practice on the weight of published orders as against published opinions was not checked. That the court lacks power to fine attorneys: the concurrence says the Rules appear to preclude further sanction and that the position is unclear, and those are different statements. That the court decided the firm-liability question: it did not, and footnote 4 refers the sanctions-authority question to the Rules Committee. That anyone will be disciplined: the referral is of the matter, for any investigation the Office of Disciplinary Counsel deems appropriate.
One presentational note for anyone quoting the caption. In the extracted text the Superior Court number 2023-CAB-005422 sits inside the caption block between the parties, which is a layout artifact of the two-column caption rather than part of any party's name.
Three things to carry away. First, the redaction route was rejected on these facts: a firm that offered to delete the fabricated citations and keep the brief was told that this would constitute effectively no sanction at all. Second, the panel said every firm attorney who signed the brief bears some responsibility, but the scope of that responsibility is expressly unsettled and has been referred to the court's Rules Committee, so it should not be reported as decided. Third, the operational guidance is in the concurrence, not the order: ban AI drafting of briefs from scratch, put cite checking on the principal drafter of each section rather than on every named attorney, use a trained paralegal for pre-filing vetting, and keep a written AI policy you can actually hand over, because the firm here did not share its policy with the court or explain how it had been communicated to its lawyers.
Source File
Open the order and confirm five things: the caption showing No. 24-CV-1099 with the panel of Easterly and Shanker, Associate Judges, and Glickman, Senior Judge; the line ORDERED that appellee's brief is stricken; footnote 5, recording the rejection of the firm's redaction proposal; footnote 4, referring the sanctions-authority question to the court's Rules Committee; and Part I of the concurrence, where Judge Glickman works through D.C. Appellate Rules 38 and 46 and concludes that the Rules appear to preclude further sanction.
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. Douglas v. Deutsche Bank National Trust Company, No. 24-CV-1099, published order filed 3 September 2026, footnote 5
FAQ
What sanction did the court impose?
It struck the appellee's brief and referred the matter to the Office of Disciplinary Counsel for any investigation that office deems appropriate. No monetary sanction was imposed. The court also referred the question of its own sanctions authority to its Rules Committee.
Why no fine?
The concurring judge, Judge Glickman, works through the available Rules and concludes that beyond the admonishment and the striking of the brief, the court's Rules and case law appear to preclude sanctioning the attorneys. He describes the position as unclear and suggests the full court may wish to address it. That is his analysis in a concurrence, not a holding of the panel.
Could the court accept the brief with the bad citations redacted?
Not in this case. The firm proposed exactly that, asking the court to take the brief as filed with the erroneous citations redacted rather than to refile it. The court refused, stating in footnote 5 that this would constitute effectively no sanction at all.
Does this make law firms liable for a colleague's hallucinated citations?
It does not decide that. The panel said every firm attorney who signed the brief bears some responsibility, and cited Superior Court Civil Rule 11(c)(1) on joint firm responsibility only as a comparison, that Rule not being a Rule of the appellate court. The concurrence agrees with the general principle but says it needs clarification, and the scope question is referred out. Treat it as open.
Was the lawyer found to have acted in bad faith?
No such finding was made. Judge Glickman writes that it does not now appear, so far as appears on the existing record, that either the attorney or the firm acted in bad faith, and that their conduct appears to have been negligent or grossly negligent rather than intentional, knowing or reckless. He adds that recklessness may be a close question given the publicity about AI in legal work, and that the factual record necessary to support such a finding has not been made. That is a statement about an undeveloped record, not an exoneration.
What should a firm do differently after this order?
The concurrence sets out the measures it considers practical: ban outright the practice of asking AI to draft a brief from scratch, hold the principal drafter of each section responsible for that section's citations rather than requiring every named attorney to check everything, use a trained paralegal for cost-effective pre-filing vetting, and maintain a written AI policy with training and confirmation of compliance. The firm in this case claimed to have such a policy but did not share it with the court.
Related briefings
Sponsored Training
Practical AI training for regulated professionals, built around verification, documentation and a defensible process. See the courses.