Federal Court Holds Certain AI‑Generated Materials Are Not Privileged
03/05/26
On February 17, 2026, the U.S. District Court for the Southern District of New York issued a ruling in United States v. Heppner, No. 25‑cr‑00503‑JSR (S.D.N.Y. Oct. 28, 2025), holding that the attorney‑client privilege and the work‑product doctrine did not protect a defendant’s artificial intelligence‑generated documents from disclosure.
Key Takeaways from United States v. Heppner
- A client’s independent use of publicly available AI to assess legal exposure will likely fall outside the protection of the attorney‑client privilege and the work product doctrine, even when the resulting materials are later shared with an attorney.
- The decision reinforces traditional privilege principles which apply regardless of the technology involved. Communications shared with third‑party platforms are not the same as communications with an attorney, may lack the confidentiality required for privilege, and materials created outside an attorney’s direction or supervision may not qualify for work‑product protection.
Background of the Case
After being charged with securities and wire fraud, the defendant independently used a publicly available generative AI platform to prepare documents analyzing potential defenses and legal arguments. He later shared those AI‑generated documents with his attorney and asserted that they were protected by the attorney‑client privilege and the work product doctrine. Defendant’s attorney acknowledged that the AI use was not directed or supervised by the attorney.
Following the defendant’s arrest, FBI agents searched his home and seized documents and electronic devices. Among the seized materials were 31 documents memorializing the defendant’s communications with the publicly available generative AI platform. At issue before the court was whether the documents were admissible or if they were protected by the attorney-client privilege or the work product doctrine.
The Court’s Rulings
No Attorney‑Client Privilege. The court held that the AI‑generated materials were not protected by the attorney‑client privilege. Applying traditional privilege principles, the court concluded that the communications failed to meet the threshold requirements because the generative AI tool was not an attorney, did not stand in an attorney‑client relationship with the defendant, was not subject to confidentiality obligations, and functioned as a third party to whom information was disclosed.
In particular, the court emphasized that the communications themselves were not confidential because the AI tool is a third party platform and disclosure to a third party generally defeats privilege unless the third party is necessary to facilitate legal advice and is subject to confidentiality obligations. The court further noted that the AI provider’s privacy policy expressly permitted the collection and use of user inputs and the AI tools’ outputs to train the AI model and to disclose such data to a host of other third parties, including governmental regulatory authorities, further undermining any claim of confidentiality. As a result, the court concluded the requisite conditions were not satisfied by the defendant’s independent use of a public AI platform.
The court’s analysis did not change just because the defendant ultimately shared the AI output with his attorney because non-privileged communications do not become privileged communications when shared with an attorney.
No Work‑Product Protection. Separately, the court also analyzed and rejected the defendant’s claims that the AI output was covered by the work product doctrine, which provides protections for materials prepared by or at the request of an attorney in anticipation of litigation or for trial. The court found that the materials were not prepared “by or on behalf of” an attorney in anticipation of litigation or trial but were instead created unilaterally by the defendant, without attorney direction. Because the defendant acted on his own when he created the AI output and the defense attorney neither requested nor supervised the creation of the AI‑generated documents, the materials did not qualify as attorney work product or as documents prepared at the attorney’s direction.
Technology‑Neutral Application of Existing Law. While the court declined to adopt any categorical rule that AI use is incompatible with privilege or work‑product protection, the court did emphasize that these tools are governed by existing, technology‑neutral, doctrines. The outcome turned on the manner of use: unsupervised, client‑initiated interaction with a publicly available AI tool, with broad data use and disclosure rights, rather than the mere involvement of AI. The court noted that attorney‑directed AI use, under appropriate confidentiality protections, could present a different analysis, but those circumstances were not present in this case.
The decision does not hold that all AI‑assisted legal work is unprotected. Rather, it clarifies that unsupervised use of publicly available AI tools by clients can undermine privilege and work‑product claims where an attorney does not direct or control that use.
Additional Takeaways
The Heppner ruling highlights that a lack of baseline understanding of how AI systems operate may heighten legal risk. While AI tools can offer many technological advantages in a variety of industries, companies and individuals alike can, and should, take specific steps to mitigate the risks of violating various confidentiality and privilege protections. The following are additional broader considerations:
- Public AI tools pose real waiver risk. If an AI tool does not function as a confidential agent facilitating legal advice on behalf of an attorney, privilege may be at risk.
- An intent to share materials with an attorney will not preserve privilege if the information is first disclosed to a third‑party (including an AI platform).
- AI providers’ terms of service, privacy policies, and data‑handling practices are likely to receive close scrutiny. In litigation, adversaries may rely on those terms to argue waivers or lack of confidentiality.
- Client‑generated AI analyses may be difficult to characterize as work product, even if prepared in anticipation of litigation, unless it can be shown that the materials were created at the attorney’s request and within a controlled, confidential process.
- Other contractual agreements may be impacted by inputting confidential information into an AI tool operated by a third party that is not contractually bound to maintain that information as confidential, even if no human reviews the data. This could void, cancel, or weaken existing trade secrets and it could violate the terms of certain non-disclosure agreements with other parties to which such confidential information is subject. This risk is more than theoretical and has potentially far-reaching consequences.
Bodman attorneys can provide guidance on this matter and other practical advice to meet your needs. To discuss these or any other legal issues affecting your organization, please contact your Bodman attorney, one of the authors of this update (Holly Hubert at 313-393-7527 or hhubert@bodmanlaw.com or Matthew Slipchuk at 313-393-7585 or mslipchuk@bodmanlaw.com), or any member of Bodman’s Artificial Intelligence Team. Bodman cannot respond to inquiries or receive information without first clearing potential conflicts with other clients. Thank you for your patience and understanding.