On February 10, 2026, the U. S. District Court for the Southern District of New York held that a criminal defendant could not claim attorney-client privilege over documents he produced using a commercially available artificial intelligence (“AI”) tool – even though he had input privileged information from his lawyers into the tool. This case is of likely interest to companies working to manage internal uses of AI tools, and also to operations of corporate Legal departments.
The Court’s decision was delivered orally during a hearing, but based on the motions filed in connection with it, appears to have proceeded as follows. After learning he was under FBI investigation, defendant Bradley Heppner hired counsel. He also started using a public version of the AI tool “Claude” (by Anthropic) to help prepare for the case (referred to here as the “AI Tool”). Heppner apparently typed prompts about the investigation into the AI Tool, which generated around 31 documents worth of AI conversations – which he saved on his devices and shared with his legal counsel.
Importantly, to generate these documents, Heppner prompted the AI Tool with at least some information he received from his attorneys, which was presumably privileged. However, Heppner’s counsel did not direct him to generate the documents, nor did counsel play any role in their creation.
The FBI came into possession of these AI-generated documents following Heppner’s arrest, when government authorities searched his home and seized the devices on which they were stored. Heppner claimed privilege over the AI-generated documents, while the government claimed no privilege existed, or had been waived.
Judge Rakoff of the S.D.N.Y. ruled that the AI-generated documents were not privileged.
- The work-product privilege did not apply because documents were not produced by an attorney or by direction of an attorney, and because they were generated using information shared by counsel in a previous meeting.
- The court also held attorney-client privilege did not apply to Heppner’s communications with the AI Tool and the resulting documents. Attorney-client privilege protects confidential communications made between a client and attorney that relate to the client’s request for legal services. Both Judge Rakoff and the government noted that Heppner’s communications with the AI Tool could not be “confidential,” as the tool’s privacy policy stated that the tool saves user inputs to train its AI and may disclose the data to “third parties” as well as “governmental regulatory authorities.”[1] Likewise, the Court seems to have agreed with the government that Heppner could not have communicated with the AI Tool for the purpose of seeking legal advice, since Anthropic’s public materials claim that the tool cannot provide legal counsel.[2]
Moving forward, businesses may consider reviewing their AI-related practices, policies and procedures to consider the potential risks to privilege this case may represent. Companies may want to revisit their internal policies on AI use in light of this decision. Further, this decision seems to encourage use of “enterprise” deployments of AI tools (as opposed to publicly-available versions), as well as careful review of contractual terms under which enterprise AI tools are procured.
For more information, Alston & Bird’s Privacy, Cybersecurity, and Data Strategy Team has extensive experience advising clients on legal issues related to emergent AI products, including U.S. and international AI legislation. Please contact us and sign up for alerts at AlstonPrivacy.com.
[1] See Mot. for Ruling in United States v. Heppner, No. 1:25-cr-00503, at 9 (SDNY Feb. 6, 2026).
[2] Id. at 8.
