So you remember last year when I asked you to please do your best not to appear in the “AI Hallucination Database”? See “Please Do Your Best Not to Appear in the ‘AI Hallucination Database’” (June 6, 2025). You do? Terrific. Because I don’t want you to end up like the now-former Assistant U.S. Attorney for the Eastern District of North Carolina, a man who now regrets his decision to use AI to draft a brief.
In general, I think a good rule of thumb is: try to live your life in such a way that no judge feels compelled to create a slide deck to display your many errors. This man either did not follow that rule or failed in doing so.
As Law360 reported this week (sub. req’d), on March 2 a judge ordered the prosecutor to show up and explain why he should not be sanctioned for violating Federal Rule 11 given that he signed a response brief that included multiple fabricated quotations and misstated the holdings of several cases. The other side had pointed out the mistakes, and the prosecutor then filed a surreply attributing them to the “inadvertent filing of an unfinalized draft document.” Because that is also not something a lawyer should do, this isn’t a great defense. But the judge suspected there was more to it, and that the “more” was called “artificial intelligence.”
At the show-cause hearing on March 10, the prosecutor confirmed that this was indeed the case.
According to the report, he told the judge he had used AI to “‘catch up’ on a draft filing after realizing he’d accidentally overwritten it.” At first I thought “overwritten” meant it was too long or too flamboyant, like the brief was in the style of Edgar Allan Poe or something. But it probably means he claimed to have lost a lot of work because he saved the wrong version. Whatever it meant, the court wasn’t buying it at all. In fact, this happened:
Judge Numbers, clearly skeptical of [the prosecutor’s] statement, pulled up a slide deck which enumerated several errors in recent case filings and asked him to explain every single one. He said the mistakes appear “diametrically opposed” to [his] statement that the errors were unintentional.
“It’s difficult to credit your response given what you’ve done here,” Judge Numbers said, referencing a misattributed quote in circuit court case law.
I have been criticized by judges, but that criticism has usually been relatively minor, and definitely has never been laid out in a slide deck so the judge and I could go through my errors one by one. I’ve had relationships that ended that way, but so far my professional career has been free of it. I hope to keep it that way.
This is yet another example of a case where someone got caught using AI and not checking the results, and then did not admit what they had done. He didn’t mention AI in the surreply at all, and when the judge asked him about that omission at the hearing, he “replied that he and [his boss] decided the surreply was appropriate”–which isn’t really an answer. He insisted that all he had done wrong was to inadvertently file an unedited draft. The judge still wasn’t buying it.
Judge Numbers didn’t let up, expressing confusion at the concept of filing anything that wasn’t final. He asked why, for any reason, [the prosecutor] used generative AI to draft anything, knowing that attorneys around the country are facing sanctions for AI use.
Because he wanted to end up in the AI Hallucination Database? Sorry, I guess that was a rhetorical question.
By the way, that database is now up to 1031 cases, 730 of them in the United States (but only one so far in Papua New Guinea, so good for them). Please don’t add to it.