Federal courts are now confronting the question of whether a client’s AI-generated documents—created using prompts that incorporate information learned from counsel—fall within the attorney-client privilege. Judge Rakoff of the Southern District of New York has ruled from the bench that they do not in the case of United States of America v. Bradley Heppner.
Prior to his arrest, Defendant Hepner ran queries about the FBI’s criminal investigation of him through an AI tool (Claude) created by a third-party company, Anthropic. The prompts included information the Defendant learned from his counsel. Defendant later shared them with his defense attorney. Prosecutors sought to use the AI documents as evidence, but Defense counsel asserted they were privileged. On February 6, 2026, prosecutors moved for a ruling on privilege, arguing three central points. First, they argued that the AI generated documents do not meet the definition of attorney-client privilege. Second, prosecutors highlighted that later transmission of the documents to defense attorneys cannot retroactively cloak unprivileged documents with privilege. Finally, prosecutors argued that work product does not shield the documents because the documents were created independently by the Defendant, not at the direction or request of his attorneys.
Judge Rakoff granted the government’s motion from the bench, memorialized only by the spare docket note: “Government’s motion granted.” The ruling may be short on words, but its consequences could reverberate widely in litigation involving AI-generated materials.
The immediate significance of this ruling is two-fold. First, lawyers should take care in cautioning clients that any divulgence of attorney-client conversations to any third party, including an AI robot, could waive the privilege. Second, litigators should consider whether to request a party’s AI conversations during the discovery process.
[View source.]
