District Court Rules Disclosures to AI Tool are Not Privileged or Confidential
District Court Rules Disclosures to AI Tool are Not Privileged or Confidential
By Hannah F. Atkinson & John H. Kester
More and more businesses and individuals are turning to public artificial intelligence (“AI”) tools like ChatGPT and Claude for analysis and strategic advice. However, reliance on these useful tools is not without risks—particularly when potentially confidential or sensitive information is shared with the AI agent. In the legal context, disclosure of otherwise privileged information to a public AI agent can result in waiver of the attorney-client privilege and the attorney-work product immunity.
The U.S. District Court for the Southern District of New York (the “Court” or “SDNY”) ruled last week that a criminal defendant’s communication with the AI tool Claude was not protected by attorney-client privilege or the work product doctrine. It noted that its decision in that matter, United States v. Heppner (S.D.N.Y. Feb. 17, 2026), appeared to be the first time “nationwide” that a court had ruled on the subject.
The attorney-client privilege protects communications between a client and his or her attorney that are: (1) for the purpose of obtaining legal advice, and (2) intended to be and are kept confidential. The work-product doctrine provides a different degree of protection for materials prepared by or at the behest of counsel in anticipation or litigation or for trial.
In rejecting the assertion of the attorney-client privilege to the defendant’s exchanges with the AI agent, the Court emphasized that the defendant “does not, and indeed could not, maintain that Claude is an attorney.” The Court additionally noted that the defendant could not reasonably believe the communications were confidential because Claude’s written privacy policy allowed for public dissemination of user inputs. Moreover, the defendant’s later sharing of his Claude exchanges with his attorney could not retroactively impart attorney-client privilege. The Court noted that “it is black-letter law that non-privileged communications are not somehow alchemically changed into privileged ones upon being shared with counsel.”
The Court similarly rejected any work-product doctrine shield for the communications with Claude because the communications were “not ‘prepared by or at the behest of counsel.”
In light of the Court’s ruling, open questions remain as to (a) how other courts will rule on similar facts in criminal and civil litigation; (b) how courts will view AI communications made by counsel; and (c) how courts will view exchanges with a private AI agent or a public AI tool with a robust privacy policy.
Until the ground rules for protection of information shared with AI agents are clarified in subsequent court decisions, businesses and individuals should adopt a conservative approach that assumes communications with AI tools are unlikely to be protected by attorney-client privilege or the work-product doctrine. Therefore, it would be prudent to:
- Avoid disclosing potentially privileged or confidential information in communications with AI tools;
- Operate under the understanding that such communications will not achieve retroactive privilege or confidentiality if subsequently shared with counsel;
- Wherever you have communicated with an AI agent, anticipate that, should you become involved in future litigation, opposing counsel will seek records of such exchanges.
GKG Law will continue to monitor related court rulings, and will keep its clients informed. Please contact us should you have any questions.
Hannah F. Atkinson hatkinson@gkglaw.com; John H. Kester jkester@gkglaw.com
