BizTech Law Blog

What is the background on the Heppner decision?

On February 17, 2026, in the case of United States of America v. Bradley Heppner, No. 25 CR. 503 (JSR), 2026 WL 436479 (S.D.N.Y. Feb. 17, 2026), the United States District Court for the Southern District of New York (the “Court”) held that documents generated by the defendant with generative artificial intelligence (“AI”) through a publicly-available AI platform were not protected by the attorney-client privilege or work product doctrine.

Whether, when an end-user prompts and communicates with a publicly available AI tool or platform in connection with a pending investigation such communications protected by attorney-client privilege or the work product doctrine, was a question of first impression nationwide. The attorney-client privilege protects from unauthorized disclosure communications between clients and their attorneys that are intended to be, and are, kept confidential for the purpose of obtaining or providing legal advice. The work product doctrine protects documentation and materials prepared by or at the behest of an attorney in anticipation of litigation or trial, or to otherwise prepare a case for a client.

Case summary: United States of America v. Bradley Heppner

In the Heppner case, the defendant has been accused of securities fraud, wire fraud, conspiracy, making false statements to auditors, and falsifying corporate records. In connection with these allegations, the defendant was arrested on November 4, 2025. The Federal Bureau of Investigation executed a search warrant of the defendant’s home and seized numerous documents and devices. Among the materials seized included approximately thirty-one (31) documents that detailed the defendant’s communications with the AI platform “Claude”, operated by the private company Anthropic.

The defendant prepared reports that outlined a defense strategy for himself. The defendant, through his attorneys, asserted privilege over these documents arguing that the information inputted into the AI platform included information the defendant had learned from his lawyers. Moreover, the defendant argued that he had created these AI-generated documents for the purpose of speaking with counsel to obtain legal advice and sharing the contents of the documents with his counsel to discuss his defense strategy. However, the defendant’s attorneys did not direct the defendant to run these searches or initiate these prompts through Claude.

As to the attorney-client privilege issue, the Court concluded that legal issues discussed between two non-attorneys cannot be protected by the privilege, and since both the defendant and the AI platform are not attorneys, this disposes of the claim of privilege. The defendant was also not found to have communicated with Claude for the purpose of obtaining legal advice because his attorneys did not direct him to utilize the AI platform. Moreover, the communications memorialized in the documents generated with AI were not confidential because the defendant communicated with a third-party, public platform to obtain such outputs. Therefore, the defendant could have no reasonable expectation of confidentiality in his communications with Claude. Despite the defendant sharing his AI-generated documents with his counsel, the Court found the communications were not “somehow alchemically changed into privileged [documents] upon being shared with counsel.” Heppner, 2026 WL 436479, at *3.

Notably, the Court’s analysis in regards to the attorney-client privilege issue was limited to the use of a publicly available AI platform that lacked any express confidentiality covenants or contractual obligations to protect user communications. It remains an open question whether the result would be different if a client used an AI service that expressly agreed to preserve the confidentiality of user inputs and outputs.

Additionally, the Court found that the work product doctrine did not apply in the Heppner case. Though the defendant argued the AI documents were prepared in anticipation of litigation, they were ultimately not generated by or at the request of his attorneys, nor did the contents of the materials reflect the defendant’s counsel’s strategy. Thus, the defendant was not acting as his counsel’s agent when he communicated with the AI platform, Claude.

Implications for attorney-client privilege and work product when clients use public AI tools

The court in Heppner expressly declined to resolve whether an attorney’s direction to a client to use a generative AI tool could give rise to protected attorney work product or communications shielded by the attorney‑client privilege. The court left open the possibility that, under different facts, a generative AI platform might be treated as an agent of counsel for privilege purposes. That unresolved question will likely become increasingly significant as attorneys explore more structured and supervised uses of AI tools in legal practice.

Additionally, the analysis in Heppner was expressly confined to instances involving publicly available AI platforms that do not include explicit confidentiality covenants or contractual obligations to safeguard user communications. The Court’s decision leaves unresolved whether the outcome might differ if a client were to use an AI service that specifically promises to maintain the confidentiality of user inputs and outputs. Express confidentiality covenants could potentially alter what constitutes a reasonable expectation of privacy and might impact the privilege analysis, especially if the service provider of the AI platform contractually promises not to disclose information to third parties. This practical issue is likely to be of significant concern for clients considering the use of AI tools and warrants close attention as the law in this area develops.

We emphasize that Heppner serves as a clear warning to our clients and their employees regarding unsupervised use of public AI platforms to evaluate, second‑guess, or refine legal advice. Inputting legal strategy, attorney communications, or case‑specific facts into publicly available generative AI tools risks destroying confidentiality and rendering those materials discoverable. The loss of attorney‑client privilege or work product protection can have severe consequences for a case, even outside of the criminal law context. While seeking a second legal opinion is entirely appropriate, Heppner underscores that doing so through public AI platforms such as OpenAI’s ChatGPT, Google’s Gemini, or Anthropic’s Claude presents substantial legal risk to end-users.

Although Heppner arose in the criminal context, its implications extend well beyond criminal proceedings. The work product doctrine and attorney‑client privilege apply across civil, regulatory, and transactional matters, making this decision relevant to litigants and counsel in all practice areas. As generative AI becomes more prevalent, Heppner highlights the importance of establishing clear policies and controls around AI use to preserve fundamental confidentiality protections, trade secrets, and defense strategies.

What are the key takeaways to avoid waiver risks when using public generative AI?

The Court’s decision in Heppner is significant because it represents an issue of first impression. As courts continue to confront questions surrounding AI‑generated content, future decisions may refine or even alter the legal framework governing privilege, waiver, and discoverability in this context. Legislative or judicial developments may ultimately impose additional safeguards, such as requiring a showing that a party knowingly and voluntarily waived applicable privileges through the use of AI tools; however, no such protections currently exist. Accordingly, both attorneys and clients should proceed on the assumption that unsupervised use of generative AI may result in an immediate and irrevocable loss of privilege.

For guidance on managing AI‑related legal risks or developing best practices in this rapidly evolving area, please connect with us via the following contact information:

• Taylor Gast...517.371.8238...tgast@fosterswift.com
• Lindsey Mead…517.371.8326...lmead@fosterswift.com