US v. Heppner, 25 Cr. 503 (SDNY): AI Use Foils Defendant’s Privilege Claim

Summary

As described below, the use of artificial intelligence models (particularly commercial versions) to assist a layperson (non-lawyer) in formulating their own legal defense strategy may not be confidential and will likely not be eligible for the attorney-client privilege or work product doctrine even if shared with an attorney after the fact.Here are a few suggestions that may enhance a claim of privilege:

The simplest and most effective approach might be to discuss your defense with your attorneys and let the legal team devise your defense strategy; avoid altogether your own use of AI, particularly on your own initiative.

Alternatively, if you do use AI on your own, commercial versions of AI that use queries to train the AI model or that have terms of service that indicate there is no expectation of privacy or that can be turned over to third parties should be avoided. The same is true of AI note takers that transcribe conversations or meetings.

Prompts of the AI tool should explicitly state that the questions asked were formulated at the request of an attorney; similarly, it would be preferable to have a written request from a formally engaged attorney requesting the AI searches to back up the claim in the prompt. 

Background

In November of 2025 the United States (“US”) Attorney’s Office for the Southern District of New York (“SDNY”) unsealed an indictment charging Bradley Heppner with, among other things, securities fraud, wire fraud, conspiracy and false statements to auditors. Judge Jed Rakoff will preside over the trial which is scheduled for April 6, 2026.Mr. Heppner was the chairman of the now bankrupt GWG Holdings, Inc. (“GWG”), a Nasdaq-listed financial services company. He is alleged to have misappropriated over $150 million from GWG through two affiliated companies that he controlled (but concealed his control): (1) Highland Consolidated Limited Partnership (“HCLP”) and (2) Beneficent Company Group, L.P. (“Beneficent”).

GWG’s resulting bankruptcy resulted in losses of over $1 billion to retail investors, predominantly retirees, that had purchased GWG L Bonds.

GWG had an equity interest in Beneficent and Heppner persuaded a special committee of GWG’s board of directors to authorize payments from GWG to Beneficent so that Beneficent could make payments on debt Beneficent purportedly owed to HCLP. Those “debt” payments ultimately went to Heppner.

The FBI arrested Mr. Heppner at his Dallas home on November 4, 2025 and executing a search warrant found and seized multiple electronic devices containing the results of research conducted by Mr. Heppner using a commercial version of Claude, the artificial intelligence (“AI”) model created by Anthropic (collectively, the “AI Documents”). The AI Documents were created by Mr. Heppner after he received a grand jury subpoena.

Claims of Privilege

Mr. Heppner argued that the AI documents should not be entered into evidence as they are subject to either the attorney-client privilege or work product doctrine because he used information he learned from his lawyers (Quinn Emanuel) as inputs for his Claude searches and he created the AI Documents for the purpose of talking to his lawyers.The US Attorney for the SDNY, Jay Clayton, filed a memorandum on February 6, 2026 with the court arguing that the AI Documents are not privileged.The Judge heard oral arguments on February 10, 2026 and ruled (in a case of first impression nationwide) that the AI Documents are neither protected by the attorney-client privilege nor shielded by the work product doctrine.

Attorney-Client Privilege

First, the AI Documents are not communications between a client and attorney and no attorney was involved when Heppner created them.Second, they were not made for the purpose of obtaining legal advice: Claude’s terms of service state that it does not provide legal advice.And, third, the documents are not confidential; the AI platform’s privacy policy permits disclosure to governmental authorities and other third parties (even in the absence of a subpoena compelling it to do so). Claude’s terms of use expressly provides that users have no expectation of privacy in their prompts which are used to train the AI model or in the model outputs.

Mr. Heppner may not subsequently transform unprivileged documents into privileged communications by later sending them to his lawyers.

Work Product Doctrine

And the work product doctrine (which is distinct from but related to the attorney-client privilege) does not protect the AI Documents because Heppner created them on his own volition and did not do so at his lawyers’ request or direction.The doctrine only shields materials prepared by or for a client’s attorney--not a non-attorney’s independent internet searches.And the AI Documents do not reflect defense counsel’s strategy.Heppner was not acting as his counsel’s agent when he communicated with Claude.

Conclusion

While the use of AI has become ubiquitous it presents a new frontier in the evolving dialogue between law and technology.Its use may be hazardous to later claims of privilege if AI searches are not confidential or are not done at the behest of or by a lawyer.This presupposes that a written and executed legal engagement letter with an attorney is in place prior to any use of an AI tool.Perhaps an enterprise (as opposed to a commercial) version of an AI tool (or a proprietary small language AI model) may give the user some expectation of privacy or confidentiality; this must be explicit in a written contract and not superseded by the AI tool’s privacy policy, constitution or terms of use. Other AI tools, such as AI note takers, may also foil claims of privilege even if lawyers are meeting participants if the resultant notes are available to governmental authorities or third parties the way Mr. Hepper’s Claude searches were.