Are AI Chats Privileged? Landmark Federal Court Ruling on GenAI & Legal Protection
In what appears to be the first federal court ruling of its kind, Judge Jed S. Rakoff of the United States District Court for the Southern District of New York has held that a defendant’s written exchanges with a generative artificial intelligence platform are protected by neither the attorney-client privilege nor the work product doctrine. The decision, issued February 17, 2026, in United States v. Bradley Heppner , carries significant implications for lawyers, clients, and anyone who turns to AI tools when navigating legal trouble.
The ruling arrives at a moment when AI platforms have become ubiquitous. The court noted that more than half of U.S. households have adopted AI in some form, and that one prominent AI platform is used by more than 800 million people worldwide each week. Yet as the court observed, the legal implications of that adoption are only beginning to be worked out.
Background: The Heppner Case
Bradley Heppner was indicted in October 2025 on charges of securities fraud, wire fraud, conspiracy, making false statements to auditors, and falsifying corporate records. The charges arose from his alleged role as an executive at GWG Holdings, Inc., a publicly traded company, where he is accused of defrauding investors of more than $150 million through undisclosed self-serving transactions.
When FBI agents executed a search warrant at Heppner’s home following his arrest, they seized approximately thirty-one documents memorializing conversations Heppner had with Claude, the generative AI platform operated by Anthropic. According to defense counsel, those conversations took place in 2025, after Heppner had received a grand jury subpoena and after it was clear that he was a target of the investigation. Counsel represented that Heppner had used Claude to prepare reports outlining possible defense strategies and the arguments he might make regarding the facts and law that the government appeared likely to charge.
Defense counsel asserted privilege over these documents, arguing that they contained information Heppner had learned from counsel, that they were created for the purpose of speaking with counsel to obtain legal advice, and that their contents were subsequently shared with counsel. The government moved for a ruling that the documents were not privileged. Judge Rakoff granted the motion.
The Court’s Analysis: Attorney-Client Privilege
The attorney-client privilege protects confidential communications between a client and counsel made for the purpose of obtaining or providing legal advice. Courts construe it narrowly because it operates as an exception to the principle that all relevant evidence should be available at trial. Judge Rakoff found that the AI documents failed to satisfy at least two, and likely all three, of its core elements.
First Element: No Attorney-Client Relationship
The most straightforward ground for denial was that Claude is not an attorney. The court cited the settled proposition that without an attorney-client relationship, a discussion of legal issues between two non-attorneys is not privileged. The court also rejected the argument that AI communications should be treated like other internet-based software tools, like word processing software. Recognized privileges require a trusting human relationship with a licensed professional who owes fiduciary duties and is subject to discipline. No such relationship can exist between a user and an AI platform.
Second Element: No Confidentiality
The court also found that the communications were not confidential. Claude’s users consent to a privacy policy under which Anthropic collects data on both user inputs and Claude’s outputs, uses that data to train Claude, and reserves the right to disclose such data to third parties, including governmental regulatory authorities. The policy expressly permits Anthropic to disclose personal data in connection with claims, disputes, or litigation, even without a subpoena compelling it to do so.
Heppner, therefore, had no reasonable expectation of confidentiality in his communications with Claude. The court observed that his AI exchanges were not analogous to confidential notes a client prepares with the intent of sharing them with an attorney, because he first shared the equivalent of those notes with a third party, the AI platform.
Third Element: Purpose of Obtaining Legal Advice
The court acknowledged that this was the closest question of the three. Defense counsel argued that Heppner used Claude for the express purpose of preparing to speak with counsel. But counsel also conceded that Heppner had not done so at counsel’s direction or suggestion. The court found that distinction decisive. Because Heppner acted on his own initiative, what mattered for privilege purposes was whether he intended to obtain legal advice from Claude itself, not whether he later shared Claude’s outputs with his lawyer.
On that question, the court pointed to Claude’s own statements. When the government asked Claude whether it could provide legal advice, Claude responded that it is not a lawyer and cannot provide formal legal advice or recommendations, and recommended that users consult a qualified attorney. The court also noted, as a matter of black-letter law, that non-privileged communications do not become privileged merely because they are later transferred to counsel.
The Court’s Analysis: Work Product Doctrine
The work product doctrine protects materials prepared by or at the direction of an attorney in anticipation of litigation, sheltering the mental processes of counsel in preparing a client’s case. The doctrine is also construed narrowly and is not lightly extended to materials prepared neither by counsel nor by counsel’s agents.
The court found that the AI documents failed on both prongs. First, they were prepared by Heppner on his own volition, not at the request of counsel, meaning Heppner was not acting as counsel’s agent. Second, while counsel conceded the documents had affected strategy going forward, they did not reflect counsel’s strategy at the time Heppner created them. Because the documents neither were prepared at counsel’s direction nor disclosed counsel’s mental processes, they did not merit work product protection.
Judge Rakoff also respectfully disagreed with a prior decision in the same district, Shih v. Petal Card, Inc., which had held that work product protection extends to materials prepared by a client without attorney direction. The court reasoned that such an expansion undermines the policy animating the doctrine, which is to preserve a zone of privacy in which a lawyer can prepare and develop legal theories and strategy with an eye toward litigation, not to protect the unilateral activities of a client who has not been guided by counsel.
What This Means for Practitioners and Clients
The Heppner decision is notable both for what it holds and for what it leaves open. Judge Rakoff specifically observed that if counsel had directed Heppner to use Claude, the outcome might have been different. In that scenario, Claude could arguably be said to have functioned in a manner akin to a highly trained professional acting as a lawyer’s agent, which existing doctrine recognizes as potentially privileged.
That observation carries an important practical lesson: the same AI tool, used in the same way, may or may not be protected depending entirely on whether its use was directed by counsel. A client who turns to AI on their own initiative to think through their legal exposure, to organize their recollections, or to draft documents they intend to share with their lawyer is not necessarily creating protected materials. The work belongs to the client, not to counsel, and the communications belong to a third-party platform that has expressly reserved the right to share them.
The confidentiality point deserves particular emphasis. The court treated Anthropic’s privacy policy as effectively eliminating any reasonable expectation of confidentiality in user inputs. Lawyers advising clients who are under investigation or who anticipate litigation should be counseling them about this reality. Communications with AI platforms are not private in any legally meaningful sense, and treating them as such could have serious consequences.
Key Takeaways for Legal Practice
AI tools used independently by a client, without attorney direction, are unlikely to attract either attorney-client privilege or work product protection. The communications are not between a client and an attorney, they are not confidential given standard AI platform privacy policies, and they are not prepared at the behest of counsel.
Where attorney direction is present, the analysis may differ. Counsel who direct clients to use AI tools as part of the legal engagement, in a manner that functions as an extension of the representation, may be able to argue that the resulting communications are protected. But that argument remains untested and the Heppner court offered only a suggestion, not a holding.
Privilege logs require careful attention. Where AI-generated documents are identified in discovery, the party asserting privilege bears the burden of establishing every element. The Heppner decision makes clear that simply asserting that materials were created in anticipation of litigation, or that they were eventually shared with counsel, will not suffice.
Client counseling must evolve. Lawyers should be advising clients who are under investigation, in litigation, or anticipating either, that their AI conversations are not confidential and may be subject to government inspection. This is not merely a risk management point. It is an obligation of competent representation in an era where AI use is pervasive.
Conclusion
Judge Rakoff concluded his opinion by observing that AI’s novelty does not place it beyond longstanding legal principles. The attorney-client privilege and the work product doctrine exist for defined purposes and require defined elements. A generative AI platform satisfies none of them when used independently by a client, regardless of how sophisticated the platform’s outputs may be or how closely the client’s intentions mirror those that might otherwise support a claim of privilege.
The legal profession is still working out the full implications of AI for the practice of law. Heppner is an important early data point. It will not be the last.
United States v. Bradley Heppner, Case No. 25 Cr. 503 (JSR), Memorandum, S.D.N.Y. Feb. 17, 2026 (Rakoff, J.). This article is provided for informational and educational purposes and does not constitute legal advice.