AI communications are not privileged
AI research by businesses and clients lacks legal protection—exposing sensitive information to discovery.
A recent court ruling makes one fact undeniable: using AI for legal research provides no shield of privilege, even when clients intend this work for counsel. Because general AI platforms offer no true expectation of privacy, anything you submit may be obtained by opposing parties as easily as any other unprotected document. Relying on AI for legal research can put your strategy—and your case—at risk.
Only your attorney can guarantee AI research remains protected by privilege. Do not expose your confidential legal strategy—let a licensed professional safeguard your interests.
In summary: If you value confidentiality and want to protect your legal rights, entrust all AI-related legal research to your attorney. Only then is your information secure, privileged, and shielded from your adversaries. If clients attempt to conduct AI research themselves, they risk compromising their own case—virtually handing sensitive information to the opposition. The stakes could not be higher.
Landmark Court Ruling: Attorney-client privilege lost when clients use AI for legal research
On February 17, 2026, Judge Jed S. Rakoff of the Southern District of New York issued a landmark decision with far-reaching consequences: written exchanges between a criminal defendant and a generative AI platform are not protected by attorney-client privilege or the work-product doctrine. United States v. Heppner, Case No. 1:25-cr-00503-JSR (S.D.N.Y. Feb. 17, 2026). This precedent should serve as a clear warning for all clients and businesses.
Explanation of the case
After receiving a grand-jury subpoena, Heppner used Anthropic’s Claude AI to generate written analyses of potential defense strategies—believing these would remain private. In reality, large language models like Claude, ChatGPT, and Google Gemini often use user queries to expand their knowledge base, and unless the highest-level, most secure subscription is used, any information entered may become publicly accessible. The risk: your legal strategy could end up in the hands of your adversaries.
The defendant later shared those AI-generated documents with counsel. The government sought a ruling that the materials were not privileged.
AI research conducted by a client—rather than by their attorney—is not protected by privilege
The court determined that the AI documents failed to meet the elements of the attorney-client privilege: communications between client and attorney, intended to be confidential, and made for the purpose of obtaining legal advice. The Heppner court focused on four key factors:
AI platforms are not attorneys. The court emphasized that the AI documents were “not communications between Heppner and his counsel” and that “Claude is not an attorney.” Because discussions of legal issues with non-attorneys are not privileged, that fact alone was dispositive.
Privilege depends on a “trusting human relationship” with a licensed professional owing fiduciary duties—and no such relationship exists with an AI platform, according to Judge Rakoff.
There is zero guarantee of confidentiality. The court found the defendant had no legitimate expectation of privacy with Claude, whose privacy policy explicitly allows collection of user data, use for AI training, and disclosure—even to government authorities. Entrusting confidential legal research to an AI platform is tantamount to publishing it.
Simply giving AI research to counsel does not make it privileged
Crucially, the court ruled that forwarding AI-generated work to your attorney after the fact does not make it privileged. Privilege must exist at the time of creation. Once exposed, your information is vulnerable—and cannot be made confidential by later legal review.
Any privilege is instantly lost the moment confidential information is shared with an AI platform—just as if you had disclosed it to a random third party. Unlike with email or secure file-sharing servers, there is no reasonable expectation of privacy. The consequence is stark: privileged information becomes fair game for discovery.
The work-product doctrine does not protect interactions with AI platforms. The court similarly rejected the defendant’s argument that the information shared with Claude was covered by the work-product doctrine, which protects materials prepared “by or at the behest of counsel in anticipation of litigation.”
Even assuming that the AI documents were created in anticipation of litigation, they still failed the doctrine’s core requirement: They were not prepared “by or at the behest of counsel.” Rather, they were prepared by the defendant “on his own volition,” and they did not reflect counsel’s strategy when they were created. The court’s ruling adheres to the general rule that the work-product doctrine exists to protect lawyers’ mental processes.
RJC – April 22, 2026
