AI Is Not Your Lawyer: Discovery Risks for Legal and IP Strategy

Using artificial intelligence (AI) platforms to assist with research, product development, or intellectual property strategy can create discoverable records that may waive confidentiality and privilege protections. In United States v. Heppner, Judge Jed S. Rakoff found that a criminal defendant’s use of the public AI platform “Claude” was not protected by the attorney-client privilege or the work-product doctrine and ordered the defendant to produce documents reflecting his interactions with the platform. Although the dispute arose in a criminal prosecution, the decision implicates risks for any individuals or organizations that rely on AI tools to analyze or address legal issues.

The Court’s Ruling

Bradley Heppner was charged with committing a crime related to an alleged investment-fraud scheme.[1] He retained defense counsel, but separately, and without direction from counsel, Mr. Heppner used an AI platform to research potential defenses, including relevant legal issues and possible defense strategies.[2] Mr. Heppner saved his prompts and the AI’s responses to later share with his defense counsel.[3] While executing a search warrant, federal investigators found documents reflecting the defendant’s use of AI for legal research.[4] Mr. Heppner’s counsel sought to withhold the AI-related materials from discovery, asserting that they contained privileged legal analysis that had been shared with counsel.[5] The court rejected this argument, finding that the documents were not privileged and could not be withheld.[6]

Relying on longstanding doctrine, the attorney-client privilege protects confidential communications between a client and an attorney made for the purpose of obtaining legal advice.[7] The court determined that the attorney-client privilege did not apply to the defendant’s AI use because Mr. Heppner’s AI exchanges did not take place with an attorney (or agent), but rather a third-party company operating an AI platform.[8]

The court also emphasized the lack of confidentiality in the AI platform.[9] Under its operating terms, the AI platform could retain and review users’ inputs.[10] The court concluded that submitting confidential information under those operating conditions constituted voluntary disclosure to a third party—disclosure of confidential information to a third party generally waives the attorney-client privilege.[11] The court further noted that even if Mr. Heppner had included advice originally obtained from counsel, once he entered that advice into AI, privilege no longer extended to those communications because the information had been shared with a third party.[12]

The court also rejected the defendant’s argument that the work-product doctrine protected the defendant’s AI usage.[13] Generally, the work-product doctrine applies to materials prepared by attorneys or their agents for the purpose of preparing for litigation.[14] The court found that because the defendant, independent of his counsel, generated the AI prompts and responses, the prompts and responses were neither drafted by counsel nor created at counsel’s request.[15] Although Mr. Heppner may have anticipated the possibility of litigation when he conducted research using AI, the court found that the documents and his usage were more akin to personal research than attorney-driven litigation preparation.[16] Thus, the court concluded that the work-product doctrine did not apply.[17]

Implications and Takeaways

While this case arose during criminal proceedings, Heppner has ramifications for civil matters, including intellectual property-related disputes. As AI usage becomes more ordinary and widespread, companies and in-house attorneys should be cautious about what information is shared with AI platforms, as those communications likely will not be protected by privilege or other applicable immunities.

In reaching its conclusions, the court focused on the fact that Mr. Heppner used a publicly accessible AI platform.[18] The platform did not keep information confidential and allowed the platform’s operators to review and possibly use submitted information.[19] The decision in Heppner may have had a different outcome if the defendant had used an AI platform with confidentiality protections. Many AI providers now offer enterprise agreements that limit operator access to, and retention of, submitted data and information, meaning that prompts entered into the platform may not be accessible to others outside the enterprise license. In that scenario, it is possible that confidentiality and privilege may be preserved. This, however, is an issue that courts have yet to address.

AI use also has implications beyond privilege and work product immunity. Engineers, researchers, and technical teams increasingly use AI tools to summarize literature, identify relevant prior art, and brainstorm potential inventions or improvements. If such AI use occurs outside the direction of counsel, the prompts entered and resulting responses could become discoverable.

Moreover, AI-generated research may play an important role when legal issues such as inequitable conduct or inventorship arise. Records reflecting AI analysis of specific patents, publications, or technical evidence could be cited as evidence that an individual was aware of certain prior art during prosecution and could support inequitable conduct allegations. Similarly, using AI to brainstorm, research, and develop technology could raise inventorship and ownership issues.

Beyond patent concerns, using AI could also jeopardize trade secret protections. If confidential trade secrets are submitted and accessed externally, the trade secret would conceivably lose its protected status.

[1] United States v. Heppner, No. 25-cr-503, 2026 WL 436479, at *1 (S.D.N.Y. Feb. 17, 2026).

[2] Id.

[3] Id.

[4] Id.

[5] Id. at *2.

[6] Id. at *1.

[7] Id. at *2.

[8] Id.

[9] Id.

[10] Id.

[11] Id.

[12] Id. at *3.

[13] Id. at *4.

[14] Id. at *3.

[15] Id. at *3–*4.

[16] Id.

[17] Id.

[18] Id. at *2.

[19] Id.