Does attorney-client privilege extend to exchanges with AI platforms? U.S. courts offer mixed messages

The attorney-client privilege applies to communications between a client and his or her attorney that are intended to be, and in fact were, kept confidential for the purpose of obtaining or providing legal advice.” The attorney work product doctrine shelters the mental processes of the attorney, providing a privileged area within which he can analyze and prepare his client’s case.

In U.S. v. Heppner, a criminal action in which the defendant is represented by counsel, U.S. District Judge Jed Rakoff determined on Feb. 27 that the AI exchanges at issue were not subject to the attorney-client privilege or attorney work product doctrine.

In Warner v. Gilbarco, a civil employment discrimination case in which the plaintiff is proceeding pro se, U.S. Magistrate Judge Anthony Patti determined on Feb. 10 that the AI exchanges at issue were protected attorney work product.

Comparing the Heppner and Gilbarco decisions: It’s all about the details

In Heppner, in arresting the defendant, the government seized materials, including documents reflecting his prompts written on the generative AI platform “Claude.”

The prompts were not written at the direction of counsel. The court determined the prompts were not privileged because (i) they were not between a client and an attorney, as Claude is not an attorney – a relationship could not exist “between an AI user and a platform such as Claude,” according to court documents. The defendant had no reasonable expectation that the prompts were confidential because Anthropic’s privacy policy expressly states such prompts may be disclosed to third parties, and the defendant could not have written the prompts to obtain legal advice because “Claude disclaims providing legal advice,” court documents said.

The fact that the defendant intended to share Claude’s responses with his counsel, and did so, did not convert them or his prompts into privileged information, the judge ruled.

The Heppner Court also determined the AI exchanges were not subject to the attorney work product doctrine because they were not conducted “by or at the behest of counsel,” and thus did not reflect counsel’s mental impressions at the time of the exchanges.

Are exchanges with ChatGPT protected under the attorney work product doctrine? The Gilbarco Court says yes. 

In contrast, in Gilbarco, where the defendants sought the production of “all documents and information concerning [the plaintiff’s] use of AI tools in connection with this litigation,” the Court held such exchanges were protected under the attorney work product doctrine because they were by a party and because disclosure to AI was not disclosure to an adversary or in a way likely to get to an adversary.

The Gilbarco Court reasoned that the plaintiff’s exchanges with ChatGPT were not a waiver of confidentiality because “ChatGPT (and other generative AI programs) are tools, not persons, even if they may have administrators somewhere in the background,” according to the ruling. The Gilbarco Court also determined the AI exchanges at issue were not relevant or proportional to the needs of the case.

Differentiating the Heppner and Gilbarco decisions

The Heppner and Gilbarco decisions appear to be the first to decide whether exchanges with AI are subject to the attorney-client privilege or attorney work product doctrine. However, there are meaningful differences between the cases that likely limit their precedential weight.

Heppner is a criminal case, and Gilbarco is a civil case. Heppner is represented by counsel, and Gilbarco is proceeding pro se. Heppner was decided by a U.S. District Court Judge, and Gilbarco was decided by a U.S. Magistrate Judge.

The relevance of the AI exchanges to the cases and their importance to the judges may also have been a factor, as the Gilbarco Court clearly believed the AI exchanges were irrelevant and an unnecessary distraction, calling the requests for their production a “fishing expedition.”

In Heppner, the defendant conceded that the exchanges were not at the behest of counsel, whereas in Gilbarco the plaintiff’s exchanges with AI were made in her pro se capacity.

Further, the Heppner Court determined there is no reasonable expectation of confidentiality in exchanging information with AI platforms whose privacy policy states disclosure may be made to third parties. In contrast, the Gilbarco Court reasoned that the plaintiff did not waive the attorney work product protection because she had no reason to believe the exchanges would be disclosed to her adversary.

What is the impact of these decisions on businesses and individuals?

While both decisions will likely be relied upon as these issues are litigated with greater frequency, their precedential value remains to be seen, particularly because questions about attorney-client privilege or attorney work product doctrine are very often significantly fact-intensive and need to be decided on a case-by-case basis. Even so, these decisions demonstrate that information exchanged with AI may be disclosed in litigation, which means that businesses and individuals should consider adopting policies intended to protect such disclosure.