D.C. appeals court strikes brief after lawyer cites four AI-generated cases that did not e - WJLA

DC court strikes brief after lawyer cites four AI-generated cases that did not exist

WASHINGTON (7News) — What began as a routine foreclosure appeal turned into a warning about one of the fastest-growing problems facing the courts: artificial intelligence producing information that looks authoritative, but isn't real.

The D.C. Court of Appeals struck Deutsche Bank's entire brief after judges discovered that four cases cited by one of the bank's attorneys did not exist.

The attorney acknowledged using artificial intelligence to help find case law and, according to the court, “did not verify the existence or accuracy of those citations before filing the brief.”

RELATED | 'No way it can be controlled': Expert warns dangerous AI deepfakes can fool anyone

The discovery comes less than a year after our reporting exposed growing concerns about AI-generated material making its way into courtrooms and the difficulty judges and lawyers face determining what is authentic.

During that investigation, D.C. Court of Appeals Chief Judge Anna Blackburne-Rigsby warned specifically about AI hallucinations and the danger of relying on technology without checking its work.

You hear about hallucinations. I think hallucinations are more likely to happen if you’re not checking behind you, I see one of the pros and cons of AI is it can make your work easier and faster on the one hand, the downside is it could also make you a little careless,” Blackburne-Rigsby said.

That is essentially what the court said happened here.

While reviewing Deutsche Bank's brief, the judges tried to locate several of the cases cited by the bank's lawyers. They could not.

One of the attorneys confirmed that four of the authorities cited in the filing did not exist.

The court identified the fake cases as Abadie v. District of Columbia, Cason v. National Consumer Cooperative Bank, Osborne v. District of Columbia and Woods v. United States.

The attorney acknowledged that she used Google's generative AI to find the citations, only later learning they were not legitimate legal authority and should not have appeared in a brief filed with the court.

She also apologized and told the court she deeply regretted the error and the burden it placed on the court and the parties.

The law firm representing Deutsche Bank, McCabe, Weisberg & Conway, LLC, told the court it was “genuinely remorseful for this failure” and said it intended to take measures to ensure that this “type of situation” never occurred again.

The court was sharply critical of the firm's response.

It noted that the firm emphasized the attorney was now a “former employee” and said the firm had shifted much of the blame to her without explaining what supervision or review of her work had taken place before the brief was filed.

The court's sanction was to strike Deutsche Bank's entire brief.

It also referred the matter to the D.C. Office of Disciplinary Counsel, which can decide whether to investigate further.

The underlying foreclosure appeal itself has not yet been decided.

The case is strikingly similar to warnings Blackburne-Rigsby made in our interview with her months earlier.

Our original investigation focused primarily on AI-generated video submitted to a court as if it were authentic evidence. Blackburne-Rigsby also described a different vulnerability: AI producing legal material that sounds convincing enough to pass until someone checks it.

AI will find you an answer and it'll sound great. I have some ability to be able to say, 'that doesn't sound right at all. Let me try to find the case that they cite. Oh, it doesn't exist,'” she said.

The court's new order shows the problem is not limited to fabricated videos or images.

It can reach directly into the legal arguments judges are asked to rely on.

Senior Judge Stephen Glickman, in a concurring opinion, called fabricated AI citations only the “tip of a perilous iceberg.”

The bigger concern, he wrote, is not necessarily the easiest AI mistake to spot.

A completely invented case can eventually be exposed by searching for it and discovering that it is not there. More subtle errors can be harder to detect — such as AI citing a real case but inaccurately describing what it says, misrepresenting facts or producing legal reasoning that appears persuasive but does not hold up.

Glickman also warned about lawyers relying on generative AI to do the thinking required to draft and refine a legal brief.

That distinction matters because the court did not say lawyers should never use AI.

The warning is about responsibility. The court was clear that while AI may make legal research faster and produce answers that sound convincing, lawyers remain responsible for verifying what they put before a court.

“We have a responsibility to know what this is, to understand how it comes to the courts and how it could derail proceedings and impact the fairness,” Blackburn-Rigsby said.