When a judge begins questioning an attorney, a certain silence descends upon the courtroom. Something heavier, not the courteous silence of everyday activities. In rooms where someone has been caught leaning a bit too heavily on a chatbot, that silence has been appearing more frequently lately.
U.S. District Judge Harold Mooty III in Huntsville, Alabama, suspended lawyer H. Gregory Harp from practicing in his court for six months last week. Four of the quotations in Harp’s brief were simply absent from the cases he cited. Harp erased his ChatGPT account when the judge requested records from it. After several days. Lawyers typically expect the bench to write in well-padded prose, but the judge’s language in the order does not. Harp “lied and destroyed evidence to cover up his AI usage,” he wrote, cautioning that anyone who believes federal judges will be lenient on this kind of matter is “gravely mistaken.”
That’s the kind of language that spreads quickly through faculty lounges at law schools and legal Slack channels. And it ought to.
The more significant change, however, occurred earlier this year in the Southern District of New York, where Judge Jed Rakoff decided that the work-product doctrine and attorney-client privilege did not apply to documents created by a defendant using a public AI tool. Before meeting with his attorneys, Bradley Heppner, the defendant, reportedly used the tool to consider his own legal exposure. Rakoff’s reasoning was straightforward, almost stubbornly so: the chatbot isn’t a lawyer, it owes you nothing, and you gave up any confidentiality you thought you had the moment you entered your private thoughts into a third-party platform.
This is the part that hurts defense lawyers. The silent framework supporting the entire defense bar has always been privilege. Because they are confident that their words will remain confidential, clients speak freely. The wall now has a hole in it. A client may be unintentionally creating the prosecution’s exhibit list if he spends a weekend at his kitchen table using a chatbot to walk through hypothetical defenses. The next time a business executive asks his lawyer, almost casually, if it’s okay to “just run something by ChatGPT first,” it’s easy to imagine how that exchange would go.
Speaking with professionals gives me the impression that the courts are proceeding more quickly than the bar is willing to acknowledge. Recently, an Oregon appeals court asked lawyers to reveal instances in which AI led to filing errors. Following the discovery of AI-generated errors in a murder case, a Georgia prosecutor was disciplined. Judges in various districts are making up their own rules, which are frequently sharper than anticipated and occasionally inconsistent. Attorneys who thought this was an issue for someone else, somewhere else, are now discovering it on their own docket.
The technology itself is not what worries defense lawyers the most. The majority of them make use of it. Quietly, many of them adore it. The difference between how attorneys actually use these tools and how courts are beginning to handle the results is the problem. When you draft a chatbot, it feels confidential. It doesn’t feel like a document, but rather a thought. Until it is referred to as a document by a judge.
It’s difficult to ignore the irony. The legal profession spent decades perfecting the privileged labels, sealed envelopes, confidentiality rituals, and the meticulous choreography of who can see what. And now all of that can be undone with a free web application used at 11 p.m. out of curiosity.
It is still unclear if the courts will soften or harden their position. One decision at a time, it is evident that the long-held beliefs about privilege, privacy, and the silent machinery of defense work are being rewritten in real time. Additionally, the lawyers who should be paying the highest fees are the ones who aren’t paying attention.
