A New York judge has ruled that content generated by artificial intelligence is not protected by attorney-client privilege or the work-product doctrine, even when shared with legal counsel, a decision issued February 17, 2026, in the case United States v. Heppner. The ruling, from Judge Jed S. Rakoff of the Southern District of New York, clarifies the legal standing of AI-generated materials in discovery and sets a precedent for how courts will approach the increasing apply of generative AI in legal workflows.
The case centered on a defendant who used a publicly available generative AI tool to analyze potential legal exposure, then shared the resulting analyses with his attorney. Federal agents subsequently seized the materials, prompting a government motion to compel production. Judge Rakoff found that because the AI-generated content was created independently, without direction from counsel, and on a platform lacking a reasonable expectation of confidentiality, it was not shielded from disclosure.
Legal experts say the decision underscores the importance of how and why AI tools are used in legal contexts. “Attorney-client privilege protects confidential communications between lawyers and clients made for the purpose of seeking or providing legal advice,” explained Julie Anne Halter, of K&L Gates LLP, in a recent analysis. “GenAI systems themselves…are not privileged by default even when legal in nature.”
The court specifically determined that the lack of confidentiality on the AI platform was a key factor. The ruling emphasized that the defendant had no reasonable expectation that his prompts and the AI’s outputs would remain private. Simply transmitting AI-generated content to an attorney did not retroactively confer privilege or protection, according to the court’s opinion.
The work-product doctrine, which protects materials prepared in anticipation of litigation, also failed to apply in this instance. The judge found that the AI-generated materials were not prepared at the direction of counsel and did not reflect the defense team’s litigation strategy. This contrasts with situations where counsel directs the use of AI to analyze claims or develop legal arguments, which could potentially qualify as work product.
A separate case, Tremblay v. OpenAI, Inc., highlighted this distinction. In that case, a court found that unused prompts and testing results generated during pre-suit investigation were protected as opinion work product, because they were created in anticipation of litigation. However, the court limited waiver to specific prompts and outputs affirmatively relied upon in the pleadings, rather than a blanket waiver of all related materials.
The rulings in Heppner and Tremblay emphasize the need for careful consideration of confidentiality and control when using generative AI tools. Experts recommend utilizing secure, enterprise-level platforms with robust data protection policies, supervising AI use closely, and clearly labeling protected materials. Organizations should also consider implementing nonwaiver agreements and addressing AI data in electronic discovery (ESI) protocols.
As the use of generative AI continues to expand in the legal field, courts are likely to focus on the nature of the AI platform, the existence of confidentiality protections, and the level of counsel’s direction and supervision when evaluating privilege claims. Litigators are advised to coordinate with e-discovery and information governance teams and to counsel clients on the risks associated with unsupervised AI use.