Your AI Conversations Are Not Privileged

By: Moish E. Peltz, Esq. and Elizabeth E. Schlissel, Esq.

KEY TAKEAWAYS 

• A New York Federal court just held that documents generated using an AI tool (Claude) and later shared with his attorneys are not protected by attorney-client privilege or work product doctrine. 

• An AI tool is not an attorney, and its terms of service may specifically disclaim any attorney-client relationship. Sharing case details with AI is legally equivalent to discussing them with a non-attorney friend. 

• Sending unprivileged AI-generated documents to your lawyer after the fact does not retroactively create privilege. 

• Clients should avoid privileged conversations with AI chatbots, and work collaboratively with their law firm to explore collaborative AI environments that operate within the attorney-client relationship. 

The New Court Ruling 

On February 10, 2026, U.S. District Judge Jed S. Rakoff of the Southern District of New York ruled in United States v. Heppner, No. 25 Cr. 503 (JSR), that 31 documents generated by the defendant using an AI tool and later shared with his defense attorneys are not protected by attorney-client privilege or the attorney work product doctrine, as first reported by Law360. 

The Defendant in this criminal case, Bradley Heppner, is the former CEO of Beneficient, a financial services company. He is charged with securities fraud, wire fraud, and making false statements to auditors in connection with an alleged scheme to defraud investors of approximately $150 million. Before his arrest, Heppner apparently used the non-enterprise version Anthropic’s Claude, a commercial AI platform, to prepare documents related to his legal situation. He then shared those documents with his defense counsel.  During the execution of a search warrant, the Government obtained several of Heppner’s electronic devices, which had the AI-generated documents on them. Heppner’s legal team tried to argue that these AI-generated documents were protected by the attorney-client privilege and work product doctrine.  

In response, the Government moved for a court ruling that the obtained “AI Documents” were not protected by the attorney-client privilege or the attorney work product doctrine. In a ruling from the bench, Judge Rakoff agreed with the Government, holding that privilege does not attach to documents created with an AI tool whose own user policy specifically states that the input and output is not confidential and there is no expectation of privacy when using the AI tool. 

Why Using AI Chat May Waive Privilege  

No Attorney-Client Privilege 

For a communication to be protected by the attorney-client privilege, the communication must be between a client and an attorney, made in confidence, and for the purpose of obtaining legal advice. AI chat does not appear to meet this test.

Not a communication with an attorney and not for the purpose of obtaining legal advice. An AI tool is not a lawyer. It holds no law license, owes no duty of loyalty or confidentiality, and is not subject to professional regulation. The terms of service for most commercial AI chat tools (including Claude) specifically advise users not to rely on their output for legal advice, advise consulting with a lawyer, and disclaim an attorney-client relationship. Courts have repeatedly held that discussing legal matters with non-attorneys does not create privilege.  

Not confidential. This is where the ruling carries its sharpest practical lesson for litigants, and perhaps is most controversial. The Defendant in Heppner voluntarily shared his prompts with a third-party commercial platform. The AI Platform’s privacy policy, in effect when Heppner used the tool, clearly states that the company collects data on prompts and outputs, and may disclose it to governmental authorities and third parties. As such, the Government asserted that there was no reasonable expectation of confidentiality and no attorney-client privilege.

No Retroactive Privilege 

Heppner’s counsel apparently argued to the Government that the defendant prepared the AI Documents to review with his counsel, and that by sending the AI Documents to his attorneys after the fact they would be privileged or protected by the attorney work product doctrine. Again, the Court apparently rejected those arguments. Courts in New York have consistently held that sending preexisting, unprivileged documents to an attorney does not miraculously transform the documents into privileged communications. As the Government in the Heppner case noted, if Heppner had conducted Google searches or checked out library books to assist with his legal case, those records would also not become privileged merely because he later discussed what he learned with his lawyer.  

No Work Product Protection 

The attorney work product doctrine protects materials prepared by or at the direction of counsel in anticipation of litigation. Here, defense counsel apparently conceded that Heppner prepared the AI Documents on his own initiative, not at counsel’s direction. That concession now seems fatal to the Defendant’s work product doctrine argument. Courts have repeatedly held that the attorney work product doctrine does not apply to materials in an attorney’s possession that were prepared neither by the attorney nor at the attorney’s behest. 

The Broader Problem with AI Privilege 

While this new Heppner ruling magnifies the risks of clients using AI tools in connection with criminal and civil cases, the underlying reasoning for the holding is not novel. The Court in Heppner applied longstanding legal principles to a defendant’s use of AI.  

While interacting with a conversational AI platform may feel private, it is not. This new Heppner case highlights the extreme risks of using open AI platforms in connection with legal matters. Before uploading or prompting an AI tool with potentially confidential or privileged information, users should pause and examine the risks in doing so.  

With the constant advent of new and improved AI tools, many users do not realize that their queries and outputs are not necessarily private, and if not directly for the purposes of interacting with legal counsel, are likely not privileged. The broader implication of the Heppner ruling is that the discoverability of AI prompts and outputs is not limited to criminal matters. Anyone using AI platforms should take heed that use of the tools (even to contemplate or answer legal queries) may have no expectation of privacy or legal privilege that attaches. When companies use AI tools in connection with strategic business decisions, employee evaluations, examining liability in workplace complaints, and examining legal matters they could be generating discoverable, unprivileged records that they never intended anyone outside of the organization to be able to receive and review.  

What Should We Do About It 

Law Firms and In-House Counsel Should Advise Clients Proactively 

If not already customary practice, attorneys need to immediately start informing clients (or employees within a company) that anything they put into an AI tool may be discoverable and should be presumed as not protected by the attorney-client privilege or the attorney work product doctrine.  In addition to counseling clients on the risks of AI in connection with legal matters, attorneys should start including statements in either their engagement letters or preservation letters outlining the risks associated with using AI tools.  Companies should adopt policies that contemplate these risks, and conduct internal training that advises employees on proper use (and non-use). It is imperative that all clients fully understand that their AI usage can be shared with their adversaries and used against them.  

Law Firms and Clients Should Embrace Collaborative AI 

If clients want to use AI to help process legal issues (and they clearly will, increasingly), they should discuss those needs with their legal counsel, and devise a way to do so inside the ambit of attorney-client privilege. Our firm is in the early stages of piloting collaborative AI workspaces that can be shared between attorney and client, where the AI interaction happens under counsel's direction and within the attorney-client relationship. Our view is that, structured properly, this would maintain privilege and change the analysis entirely. Please feel free to reach out if you are interested in learning more, or being a pilot user with us. 

If you have any questions about how to safely use AI platforms without waiving the attorney-client privilege, please consult with your attorney at Falcon Rappaport & Berkman LLP.