IP Hot Topic: Think Before You Prompt – Why Your AI Chatbot Conversations May End Up in Court

A federal court in the Southern District of New York recently ruled on a question of first impression that merits the attention of every business that uses artificial intelligence (AI) tools: do communications with a publicly available AI platform in connection with litigation enjoy the protection of the attorney-client privilege or the work product doctrine? The answer, according to Judge Jed S. Rakoff in United States v. Heppner, is no. While case law in this area is nascent, the ruling cautions anyone who has used an AI chatbot to think through legal strategy, draft documents, or explore potential litigation positions: those conversations may not be protected, and they could be obtained by adverse parties and law enforcement. As we explain below, a permissive approach to employee use of public AI tools carries real litigation risk.

Background: The Heppner Case

The Heppner case concerns AI-use by Bradley Heppner, a corporate executive charged with securities fraud and related offenses. When FBI agents executed a search warrant at his home, they seized documents (“AI Documents”) memorializing his conversations with Claude, the generative AI platform operated by Anthropic.

According to Heppner’s counsel, these AI Documents were created after Heppner had received a grand jury subpoena and after it was clear that the government viewed him as an investigation target. His counsel asserted attorney-client privilege and work product doctrine protection over these documents, arguing that Heppner used Claude to organize his thinking, outline potential defense strategies, and prepare materials he intended to share with his attorneys—all in anticipation of the indictment that followed. Heppner’s counsel asserted that Heppner had: (1) inputted information received from counsel, (2) created the AI Documents for the purpose of speaking with counsel to obtain legal advice, and (3) shared the contents of those documents with his counsel.

But Heppner’s counsel did not direct him to run Claude searches. The government moved to compel production of the AI Documents, arguing they were protected by neither the attorney-client privilege nor the work product doctrine. On February 10, 2026, Judge Rakoff granted the motion from the bench and issued a written opinion explaining his reasoning.

Why the Attorney-Client Privilege Did Not Apply

The attorney-client privilege protects confidential communications between a client and an attorney, made for the purpose of obtaining or providing legal advice. Judge Rakoff found that the AI Documents failed to satisfy at least two, and arguably all three, of these requirements.

First, “Claude is not an attorney.” Heppner did not contend that Claude was his attorney, and the court added that he could not contend otherwise. Because there was no attorney-client relationship between Heppner and Claude, the foundational predicate for the privilege was simply absent.

Second, the communications were not confidential. This conclusion followed directly from Anthropic’s own privacy policy, which users accept when they use Claude. That policy makes clear that Anthropic collects data on both user inputs and AI outputs, uses that data for training purposes, and reserves the right to disclose such data to a host of third parties, including governmental regulatory authorities. The court held that this policy put Heppner on notice that his conversations could be shared. He had no reasonable expectation of confidentiality in those exchanges.

Third, Heppner did not use Claude for the purpose of obtaining legal advice. The court left open a narrow question: had counsel directed Heppner to use Claude as a research or drafting tool, Claude might arguably be treated as operating in a manner analogous to a lawyer’s agent, whose communications can fall within the privilege. But because Heppner acted on his own initiative, the relevant question was whether he sought legal advice from Claude itself. And Claude, for its part, disclaims providing legal advice. Sharing those communications afterwards with Heppner’s counsel did not change the documents into privileged ones.

Why the Work Product Doctrine Did Not Apply

The work product doctrine provides qualified protection for materials prepared by or at the direction of counsel in anticipation of litigation. The AI Documents failed this standard for two independent reasons.

First, they were not prepared at the behest of counsel. Heppner’s counsel confirmed as much at oral argument. Because Heppner was acting on his own volition rather than as his counsel’s agent, the documents fell outside the scope of the doctrine as consistently interpreted by the Second Circuit.

Second, the AI Documents did not reflect defense counsel’s strategy at the time they were created. Counsel acknowledged that while the documents may have eventually influenced strategy going forward, they did not capture counsel’s thinking as it existed when Heppner generated them.

Five Takeaways for Your Business

1. AI Conversations with Public Platforms Are Discoverable.

If company employees use a publicly available AI tool—Claude, ChatGPT, Gemini, or any comparable platform—to think through legal problems, draft strategy memos, or analyze your litigation position, those conversations risk not being protected. If other courts rule the same way as Judge Rakoff, such conversations could be obtained by adversaries in civil litigation and by the government in criminal or regulatory proceedings. Treat any input you type into a commercial AI platform the same way you would treat a communication sent to a third party with no expectation of confidentiality.

2. The Attorney-Direction Rule Creates an Important Planning Opportunity.

Judge Rakoff specifically noted that the outcome might be different had counsel directed Heppner to use Claude. When AI tools are used at the explicit direction and supervision of counsel—as a component of the attorney’s own analytical or drafting process—there is a stronger argument that the resulting materials reflect the attorney’s mental processes and may warrant protection. But international and foreign companies should note that “not every communication between any client and foreign attorney will be entitled to the privilege in U.S. courts.” Mass Engineered Design, Inc. v. Ergotron, Inc., No. 2:06-272, 2008 WL 11348359, at *3 (E.D. Tex. Oct. 14, 2008). Companies should work with litigation counsel to establish protocols that maximize the prospect of protection when AI tools are used in connection with anticipated or pending disputes.

3. Privacy Policies Destroy Confidentiality Expectations.

The court’s confidentiality analysis turned heavily on the platform’s privacy policy. Before using any AI tool in connection with sensitive business matters, legal strategy, or pending investigations, review the platform’s terms. Enterprise or API versions of many AI platforms offer different data handling commitments than consumer-facing products. Understanding these distinctions—and selecting appropriate tools accordingly—can make a material difference in how a court assesses confidentiality.

4. A Clear AI Policy Was Needed Yesterday.

Heppner is just the first datapoint on this issue. As AI-related issues arise in more cases, courts may elaborate on this or other approaches to privilege. Importantly, the risks highlighted by Heppner do not arise only in the courtroom. They arise daily, in the ordinary course, every time an employee types a sensitive question into an AI chatbot. A robust AI usage policy may require:

• above all, educating employees that AI conversations may not be private, may not be privileged, and may be potentially subject to discovery;
• identifying which AI tools are approved for use in connection with litigation-sensitive matters;
• distinguishing between enterprise tools with appropriate data handling and consumer products with broad third-party disclosure rights;
• requiring that any AI-assisted work product on legal matters be conducted at the direction of counsel, and documented as such;
• establishing litigation hold procedures that are triggered at the earliest moment litigation is reasonably anticipated, and that expressly cover AI-generated documents and conversation logs; and

Companies that treat AI governance as a compliance checkbox rather than a litigation-readiness imperative do so at significant legal risk.