In United States v. Heppner, No. 1:25-cr-00503 (S.D.N.Y.), Judge Jed S. Rakoff ruled from the bench that documents prepared by a criminal defendant using a commercial AI platform are not protected by the attorney-client privilege or work product doctrine, even though the defendant shared those materials with his attorneys. The decision represents one of the first judicial rulings directly addressing privilege claims in the context of generative AI usage.
Background of the Case
According to court filings and reporting, the defendant Heppner, after learning that he was the subject of a criminal investigation and after retaining counsel, used Anthopic’s regenerative AI chat tool – Claude – to analyze his legal exposure and prepare documents outlining potential defenses and legal arguments. Heppner later transmitted these AI-generated materials to his attorneys. In addition, Heppner’s prompts included information obtained from his counsel.
Defense counsel asserted that approximately thirty-one documents, seized by the government under subpoena, were privileged materials. The documents generated by the AI tool reflected Heppner’s prompts and the AI tool’s responses. The government moved for a ruling that the materials were not protected by either the attorney–client privilege or the work product doctrine. The defense argued that the documents were prepared for the purpose of seeking legal advice and incorporated information learned from counsel.
Judge Rakoff rejected those arguments and concluded that the materials were not protected by either privilege.
The Court’s Reasoning
1. Communications Were Not with Counsel
Attorney–client privilege protects confidential communications between a client and an attorney made for the purpose of obtaining legal advice. The court emphasized that an AI platform is not an attorney and does not form part of the privileged relationship. Communications with an AI system therefore do not satisfy a core element of privilege.
2. Lack of Confidentiality
The court placed significant weight on the Anthropic’s privacy policy which provides that user inputs were not confidential. Because privilege requires a reasonable expectation of confidentiality, disclosure to a third-party system defeated the claim of privilege. In fact, its privacy policy indicates that Anthropic collects data on the “prompts” entered and “outputs” generated; that it uses this data to “train” its AI tool; and that it may disclose this data to “governmental regulatory authorities” and “third parties.”
3. Later Transmission to Counsel Does Not Create Privilege
The court also rejected the argument that documents became privileged once shared with attorneys. Courts have long held that preexisting, non-privileged materials do not become privileged simply because they are later transmitted to counsel.
4. Work Product Protection Did Not Apply
The work product doctrine protects materials prepared by or at the direction of counsel in anticipation of litigation. Here, the defendant created the AI documents independently, without direction from counsel, and the court concluded that independent AI-assisted research by a client did not qualify.
Why the Decision Matters
Although Heppner is a criminal case, its implications extend well beyond that context. Businesses increasingly use generative AI tools to analyze legal risk, draft agreements, or evaluate disputes before consulting counsel. The ruling reinforces several practical points:
AI Conversations May Be Discoverable
Information shared with AI platforms may become discoverable in litigation. As AI usage expands, courts and litigants are increasingly treating AI chat histories and generated documents as potentially discoverable materials.
Privilege May Be Lost Even When Legal Advice Is Involved
The decision underscores that privilege depends on confidentiality. Sharing otherwise privileged information with a third-party AI system may waive privilege protections, even where the ultimate purpose is to discuss the information with counsel.
AI Does Not Replace Attorney Direction
The ruling highlights the continued importance of attorney involvement where privilege or work product protection is expected. Materials prepared independently by a client using AI may not receive the same protection as materials prepared with legal counsel.
Unexpected Consequences
The content of a chat prompt may be direct or circumstantial evidence of the inquirer’s intent to infringe leading to awards of treble damages and attorney’s fees. Facts in the prompt that are incomplete, misunderstood, or incorporate or rely on AI hallucinations, though wrong, may be admitted as adverse evidence against the inquirer’s business. Simple inquiries about inventorship, patent infringement, patent validity, or prior art may be adversely construed. Premature disclosure of inventions may destroy novelty. Beyond litigation considerations, contracts prepared without legal overview may omit essential provisions and include terms that are unfavorable, inapplicable, and unnecessarily impede negotiations.
Takeaways
For businesses engaged in potential or actual litigation, product development, patenting, or intellectual property enforcement, the implications may be particularly significant. Company employees who use AI tools to analyze questions involving litigation, inventorship, patentability, patent novelty, validity or infringement can expose a business to unanticipated risks. Statements made in those contexts, even if exploratory or incorrect, may later be found to be admissions or evidence of knowledge or wrongful intent if they become discoverable.
Additionally, disclosure of technical or invention-related information through AI platforms may raise separate concerns regarding confidentiality and, in absolute novelty jurisdictions outside of the U.S., result in the loss of patent rights.
The Heppner decision does not create a new category of privilege law; rather, it applies traditional privilege principles to modern technology. Courts are likely to continue addressing similar issues as AI becomes more integrated into business and legal workflows.
For now, the decision serves as a reminder that generative AI tools are not confidential advisors. Unless directed by counsel, business communications involving legal strategy or sensitive business information should be provided to and managed by legal counsel.