AI-Generated Fake Cases Lead to Court Rebuke in Bankruptcy Dispute

In a recent issue of PC Gamer, Joshua Wolens explores the risk of using generative AI as a tool in legal practice, regarding AI-generated fake cases in a bankruptcy dispute.

On July 18, Judge Michael B. Slade of the US Bankruptcy Court of Northern Illinois found that attorney Thomas Nield and his firm violated a Federal Rule of Bankruptcy Procedure. The case is In re Chapter 13 Marla C. Martin Debtor.

Judge Slade said that he appreciated Nield’s remorse and candor after the problems came to light, but chastised him nevertheless for a Chapter 13 repayment plan that relied on ChatGPT-generated legal citations that were fabricated.

The underlying dispute dates back to 2024, when Nield’s client proposed a repayment plan to address debts owed to Corona Investments LLC.

When Corona filed an objection disputing, among other issues, the plan’s feasibility, Nield’s response argued that Corona lacked standing to raise such a challenge.

To support this argument, Nield submitted four bankruptcy cases purportedly addressing the point. However, scrutiny by Judge Slade’s chambers revealed significant discrepancies.

Three of the cases were wrongly attributed or contained quotations absent from the actual opinions, and the fourth case, In re Jager, did not exist at all.

Judge Slade’s memorandum detailed the extent of the errors. In each instance, the quoted language was missing from the cited opinions, and in some cases, the cases were from entirely different jurisdictions than claimed.

The court concluded that the “authority” in the brief was fabricated, tracing the problem to Nield’s admitted reliance on AI to produce citations. Nield conceded he had “run it through AI to some extent” without realizing the citations were false.

“I assumed that an AI program would not fabricate quotes entirely,” said Nield.

The incident underscores yet again the importance of rigorous verification before submitting any AI-generated content to a court. AI tools can aid research, but their outputs must be cross-checked against primary legal sources to avoid ethical violations and potential sanctions.