When AI Isn’t Privileged: Litigation Strategy Risks After United States v. Heppner

By Laurie Choi and John Delikanakis

With the widespread adoption of generative AI in everyday life, it is no surprise that AI is increasingly being used by individuals and companies to address legal questions and strategy. A recent opinion from the U.S. District Court for the Southern District of New York highlights that, in the context of litigation, exchanges with AI to develop legal strategy may be discoverable and subject to disclosure.

Judge Rakoff’s February 19, 2026 opinion addressed (apparently for the first time in the United States) whether a client’s prompts to, and responses by, an open AI system can be protected by attorney-client or attorney work product privilege, and held (1) there could be no attorney-client privilege over these exchanges and (2) that, in this case, because the prompts were not directed by the attorney or reflective of the attorney’s strategy, the exchanges were not attorney work product.

In October of 2025, Bradley Heppner was indicted by a grand jury with securities fraud, wire fraud, conspiracy to commit securities fraud and wire fraud, making false statements to auditors, and falsifying corporate records. Heppner was arrested and various documents and electronic devices were subsequently seized from Heppner’s home, including documents containing communications between Heppner and generative AI platform Claude. The exchanges with Claude were not made at the direction of Heppner’s attorney but were made after it was clear that Heppner was the target of investigation by the government. Heppner’s attorney conceded that Heppner’s prompts did not disclose his strategy but argued that Claude’s response outlined arguments that might be part of his defense strategy.

The opinion applied longstanding common law essential elements for the recognition of an attorney-client relationship — communications (1) between a client and his or her attorney (2) intended to be, and in fact were, confidential and (3) for the purpose of obtaining or providing legal advice. Judge Rakoff held that the exchanges were not protected by attorney-client privilege as (1) Claude is not human, and “all recognized privileges require, among other things, a trusting human relationship,” (2) Claude is, by its nature and design, an open AI system and thus not confidential, and (3) Heppner did not use Claude at his counsel, and thus, Claude was not “function[ing] in a manner akin to a highly trained professional who may act as a lawyer’s agent within the protection of the attorney-client privilege.”

As to attorney work product privilege, Judge Rakoff held that because the exchanges with Claude were not at counsel’s direction, and because Heppner’s prompts did not disclose his counsel’s strategy, there could be no protection under the attorney work product doctrine.

The takeaway here is simple. While it may seem convenient and tempting for clients to get ahead and prepare legal analysis and strategy through readily available AI platforms, this comes with serious risk of creating sensitive documents that may be subject to disclosure. In the context of preparation for possible litigation, the best practice is to only use AI in consultation with an attorney or use AI with discretion and full understanding of the risks involved.