Day on Torts
In a memorandum opinion that every trial lawyer should read, Judge Jed S. Rakoff of the United States District Court for the Southern District of New York has ruled that a criminal defendant’s written exchanges with the generative AI platform Claude are protected by neither the attorney-client privilege nor the work product doctrine. The decision appears to be the first of its kind nationwide, and its reasoning carries significant implications for how lawyers counsel clients — and for how clients use AI tools — in the shadow of litigation.
The Facts
In United States v. Heppner, No. 25 Cr. 503 (S.D.N.Y. Feb. 17, 2026), Bradley Heppner was indicted in October 2025 on charges of securities fraud, wire fraud, conspiracy, making false statements to auditors, and falsifying corporate records arising from his alleged conduct as an executive of GWG Holdings, Inc. Following his arrest, FBI agents executed a search warrant at his home and seized, among other materials, approximately thirty-one documents memorializing communications Heppner had with Claude, the AI platform operated by Anthropic. These exchanges — which Heppner’s counsel described as “reports” outlining potential defense strategy and legal arguments — were prepared by Heppner on his own initiative, after he had received a grand jury subpoena and it was clear he was a target of the investigation. Counsel conceded, critically, that they had not directed Heppner to use Claude.
Heppner’s counsel asserted privilege over these documents, arguing that (1) Heppner had inputted into Claude information he had learned from counsel; (2) he had created the AI documents in order to prepare for conversations with counsel; and (3) he had subsequently shared the documents with counsel. The Government moved for a ruling that the documents were protected by neither the attorney-client privilege nor the work product doctrine. Judge Rakoff agreed, from the bench and now in written form.
The Attorney-Client Privilege Analysis
Judge Rakoff concluded that the AI documents failed to satisfy at least two — and likely all three — elements of the attorney-client privilege as articulated in United States v. Mejia, 655 F.3d 126 (2d Cir. 2011): that the communications be (1) between a client and counsel, (2) kept confidential, and (3) made for the purpose of obtaining or providing legal advice.
On the first element, the court’s analysis was swift and sound: Claude is not an attorney, and Heppner did not claim otherwise. The discussion of legal issues between two non-attorneys simply does not attract the privilege. The court also rejected the argument — advanced by some commentators — that AI inputs are more akin to the use of cloud-based word processing software than to communications, noting that this framing only underscores the absence of privilege because recognized privileges require a trusting human relationship with a licensed professional who owes fiduciary duties. No such relationship exists, or could exist, between a user and an AI platform.
On the second element, confidentiality, the court pointed to Anthropic’s written privacy policy — to which users consent — which discloses that Anthropic collects user inputs and AI outputs, uses such data to train its models, and reserves the right to disclose such data to third parties, including governmental regulatory authorities. By communicating through a platform operating under such a policy, Heppner had no reasonable expectation of confidentiality. The court further noted that Heppner’s act of sharing his notes with Claude — a third party — distinguished his situation from that of a client who prepares confidential notes for the purpose of sharing them with counsel.
On the third element, purpose, the court acknowledged a somewhat closer question. Heppner’s counsel argued that he used Claude expressly to prepare for conversations with his attorneys. But the court focused on the decisive fact that counsel had not directed Heppner to use Claude. Had counsel done so, the court suggested, Claude might arguably have functioned as a professional agent of counsel, bringing the communications within the privilege’s ambit. Because Heppner acted on his own initiative, the relevant inquiry was whether he sought legal advice from Claude — and Claude’s own disclaimer that it is not a lawyer and cannot provide legal advice resolved that question against Heppner.
The court also firmly rejected the contention that subsequent disclosure to counsel could launder otherwise non-privileged communications into privileged ones. That proposition, the court observed, is black-letter law: non-privileged documents do not become privileged merely because they are later shared with an attorney.
The Work Product Doctrine Analysis
The work product doctrine presents a distinct but related set of issues. At its core, the doctrine shelters the mental processes of counsel and provides qualified protection for materials prepared by or at the behest of counsel in anticipation of litigation. Judge Rakoff held that the AI documents fail on both counts.
First, the documents were not prepared at counsel’s direction — indeed, counsel confirmed as much. Heppner acted on his own volition, and therefore was not functioning as counsel’s agent when he created the documents. Second, while counsel conceded that the AI documents ultimately influenced strategy going forward, they did not reflect counsel’s strategy at the time of their creation. The doctrine’s purpose — preserving the zone of privacy within which a lawyer develops legal theories and strategy — is not served by protecting self-initiated client communications with a third-party AI platform.
The court also expressly disagreed with Shih v. Petal Card, Inc., 565 F. Supp. 3d 557 (S.D.N.Y. 2021), which had held that the work product doctrine is not limited to materials prepared by or at the direction of counsel. Judge Rakoff found that reading inconsistent with the doctrine’s animating purpose, which the Second Circuit has repeatedly emphasized is to protect the mental processes of attorneys, not clients acting independently.
This is the first opinion of its type and one worthy of careful consideration. Plaintiff personal injury lawyers may wish to advise clients not to use AI for legal research about their case.