[EDRM Editor’s Note: The opinions and positions are those of Michael Berman.]

U.S. v. Heppner, __ F. Supp. 3d __, 2026 WL 436479 (S.D.N.Y. Feb. 17, 2026)(Rakoff J.), has exploded¹ across the blogsphere with its important, novel, and fact-based holding that a client’s artificial intelligence research: 1) was not at the direction of counsel; 2) used a non-confidential A.I. program that did not provide legal advice; and, 3) was not privileged or work product. 

I have been no exception to the interest in Heppner and blogged about the decision in A.I. Documents Deemed Not Privileged (Feb. 12, 2026). The Heppner court relied in part on an article by Prof. Ira P. Robbins, which I described in “Against an AI Privilege” – Are Prompts Discoverable?  Is Output? (Jan. 2, 2026).

So, I wondered how the Heppner court became aware of what Mr. Heppner’s attorney did or did not communicate to Mr. Heppner about the use by Mr. Heppner of A.I.? How did the court know that Mr. Heppner’s attorney did not direct Mr. Heppner to use A.I.? For reasons set out below, I suggest that ECF 23-5 was crucial.

THE HEPPNER COURT RELIED ON ECF 23-5 & THE ABSENCE OF A DIRECTIVE FROM HEPPNER’S ATTORNEY TO HEPPNER

In my view, two facts were critical in Heppner. First, the A.I. tool expressly stated that it was not confidential and did not provide legal advice. Second, the court wrote:

Without any suggestion from [defense] counsel that he do so, Heppner “prepared reports that outlined defense strategy, that outlined what he might argue with respect to the facts and the law that [the defense] anticipated that the government might be charging.” … Thus, counsel asserted, Mr. Heppner “was preparing these reports in anticipation of a potential indictment.” [emphasis added].

Id. at *1.

In “exchanges with the Government” Heppner asserted privilege. Heppner had inputted information received from counsel and created A.I. documents for the purpose of speaking with counsel. The court added:

Heppner’s counsel conceded, however, that counsel “did not direct [Heppner] to run Claude searches.” ECF No. 23-5. [emphasis added].

The Heppner court wrote that “Heppner’s counsel asserts that Heppner communicated with Claude for the ‘express purpose of talking to counsel.’ ECF No. 23-5. But, as Heppner’s counsel also conceded, Heppner did not do so at the suggestion or direction of counsel. See id. (noting that counsel ‘did not direct [Mr. Heppner] to run Claude searches’).” [emphasis added]. As to work product, the court wrote:

Heppner’s counsel confirmed that the AI Documents “were prepared by the defendant on his own volition.” Tr. at 5.² That means that Heppner was not acting as his counsel’s agent when he communicated with Claude. As to the latter, counsel conceded that while the AI Documents did “affect” counsel’s strategy going forward, they did not “reflect” counsel’s strategy at the time that Heppner created them. See id.

Id. at *3 (emphasis added). 

Thus, in the court’s words:

Here, there is no dispute that Heppner acted on his own when he created the AI Documents. See ECF No. 23-5 (noting that defense counsel ‘did not direct [Mr. Heppner] to run Claude searches’); Tr. at 5 (confirming same). Because the AI Documents were not prepared at the behest of counsel and did not disclose counsel’s strategy, they do not merit protection as work product.

Id. at *4.

THE GOVERNMENT’S MOTION CHALLENGING PRIVILEGE

While I have not read the entire record in Heppner, the A.I. issue seems to have been framed in ECF No. 22, the Government’s “Motion for a Ruling that Documents the Defendant Generated Through an Artificial Intelligence Tool Are Not Privileged.” The Government’s flagship argument on attorney-client privilege was, in summary:

[T]he AI Documents fail every element of the attorney-client privilege. They are not communications between a client and attorney—the AI tool is plainly not an attorney, and no attorney was involved when he created the documents. They were not made for the purpose of obtaining legal advice—the AI platform’s terms of service expressly disclaim any attorney-client relationship and state that the tool does not provide legal advice. And they are not confidential—the defendant voluntarily shared his queries with the AI tool, and the AI responses were generated from a third-party commercial platform whose privacy policy permits disclosure to governmental authorities.

The Government’s flagship argument on work product was, in summary:

[T]he work product doctrine does not protect these materials. Defense counsel has represented that the defendant created the AI Documents on his own initiative—not at counsel’s behest or direction. The doctrine shields materials prepared by or for a party’s attorney or representative; it does not protect a layperson’s independent internet research. [emphasis added].

The Government recited the following facts. Mr. Heppner was arrested on fraud charges. Devices were seized pursuant to a search warrant. A privilege protocol was established and the AI documents were segregated. Defense counsel provided a privilege log. The Government’s motion recites:

The Government has continued to engage with defense counsel on the AI Documents. In doing so, defense counsel informed the Government that the defendant included as inputs for the AI Documents, among other things, information the defendant learned from his counsel. (Rothman Decl. Ex. E)….  However, defense counsel has not claimed that the defendant used the AI tool or prepared the AI Documents at counsel’s direction. (See Rothman Decl. Ex. E (“[counsel] did not direct [the defendant] to run Claude searches”)). Nor has counsel claimed any other involvement in the creation of the AI Documents. [emphasis added].

In an oversimplified summary, the Government argued: “The defendant has not come close to satisfying his burden in establishing that the AI Documents are privileged. The defendant appears to have directed legal and factual prompts at an AI tool, not his attorneys. Defense counsel has informed the Government that, once the defendant obtained responses to his AI prompts, he transmitted those responses to counsel. But that act of transmission does not transform the defendant’s non-privileged use of an AI tool into a privileged communication shielded from disclosure.” 

The Government added: “But, as defense counsel has conceded in its discussions with the Government, the defendant’s AI-research was not prepared at the behest of anyone, including counsel…. Where, as here, the defendant acted alone and then transmitted the materials to his counsel, the fruits of the defendant’s actions are not attorney work product.” [emphasis in original].

THE ROTHMAN DECLARATION AND ECF 23-5

This review of ECF 22, the Government’s motion, leads us to ECF 23, the documents relied on by the court. The Rothman Declaration and ECF 23-5 were cited by the court.

Document 23 is a Declaration by Assistant U.S. Attorney Alexandra N. Rothman. It authenticates six exhibits. Exhibit A is a copy of the Privilege Protocol Stipulation. Exhibit B is a copy of defendant’s privilege log. Exhibit C is an email and privilege prepared by defense counsel. Exhibit D “is a true and correct copy of notes of a call with defense counsel….” Exhibit E is the same as Exhibit D, however, on a different date. Exhibit F is “a true and correct copy of Claude’s response when asked, ‘Can Claude provide legal advice?’”

Exhibit D, ECF 23-4, is a partially redacted email from the AUSA stating: “Claude Docs: Heppner would input facts into AI in order to provide response to attorneys for legal advice. For purpose of discussing the issues with Quinn.” Presumably, Quinn is the law firm of Quinn Emanuel Urquhart & Sullivan, LLP. The addressee is redacted.

Exhibit E, ECF 23-5, is a partially redacted email from the AUSA. It states: “2026_2_2 – tc with Quinn re Heppner.” “AR” – presumably AUSA Alexandra Rothman – raised three issues with defense counsel, Quinn. The first is “Claude AI searches.” All of the other issues are redacted. The email states that “Ben Oneil”³ said the following:

Claude AI: QE⁴ view is Claude AI searches are privileged. QE talked with Heppner re case; if BH wanted to use Claude related to case, he would put in facts things gov’t alleging, things QE told him, and would prepare reports for purpose of discussing with counsel; vehicle to consolidate thoughts; every report was created for express purpose of talking to counsel; every report based on information counsel discussed with him in first instance b/c reports pre-indictment; QE did not direct BH [presumably Mr. Heppner] to run Claude searches [emphasis added].

Exhibit 5, ECF 23-6, is a screen shot of a prompt to, and output from, Claude.

ADDITIONAL INFORMATION

In Client’s AI Chats Aren’t Privileged. What’s the Rule for Counsel? (Feb. 26, 2026), Joel Cohen and Dougals Nadjari report:

During oral argument, Heppner’s counsel said the reports weren’t prepared by him—nor with his blessing, encouragement, or approval. In fact, counsel said he was unaware that Heppner was creating these documents by “communicating” with an AI platform.

“Heppner claimed he spoke with his counsel and then generated approximately 30 documents containing his prompts and Claude’s responses (the AI Documents), which he later shared with his defense attorneys.” Federal Court Rules Some AI Chats Are Not Protected by Legal Privilege: What It Means For You | Crowell & Moring LLP (Feb. 18, 2026).

Further, “Defense counsel noted that the defendant began querying Claude after receiving a grand jury subpoena and after discussions with the government made clear that he was the target of an investigation.” S.D.N.Y. Court Considers Whether AI-Generated Documents Are Subject to Privilege Protections | Paul, Weiss (Feb. 20, 2026).

DISCUSSION

I do not practice criminal law and never have. This factual description is not a criticism of anyone. 

It appears that, during back and forth negotiations with the Government, defense counsel discussed a privilege protocol with the Government and made certain assertions to support the claims of privilege and work product protection. For example, defense counsel apparently told the Government that Mr. Heppner input “things QE [Quinn Emmanuel] told him” and that the A.I. platform “would prepare reports for purpose of discussing with counsel….” See ECF 23-5.

Apparently, more was said. The AUSA wrote that: “QE did not direct BH to run Claude searches….” The assertion by defense counsel, if accurately recorded (and there is no reason to think otherwise), may have been necessary. I do not suggest otherwise. Further, date stamps or other evidence may have made that conclusion inevitable. And, the converse –defense counsel stating that defendant used A.I. at defense counsel’s direction—would certainly be proper.

The absence of a directive—from the attorney to the client—to use A.I., does, however, appear to be one important consideration in the work product ruling. For example, one blog states:

The court considered, but did not decide, whether the outcome might differ if the defendant’s counsel had directed him to use the AI tool, potentially treating the tool akin to a highly trained expert acting as an agent of the attorney, and thus covered by the privilege. Judge Rakoff allowed that this “might arguably” be true, but that these were not the facts present in the case. He also noted that, if prompted for legal advice, the tool will apparently reply that it is not a lawyer and cannot provide such advice, and that users should consult with qualified attorneys instead. (The authors agree wholeheartedly with the AI tool).

Andrew Peck, et al., Are AI-generated documents protected from discovery if you send them to your lawyer? One judge says “no” | DLA Piper (Feb. 18, 2026).

Notes

1. Update: Judge Rakoff Issues Written Opinion that AI-Generated Documents Are Not Protected by Privilege – Debevoise Data Blog; AI, Privilege, and the Heppner Ruling: What the Court Actually Held—And How to Structure AI Use Safely | Insights | Venable LLP; Conversations with AI Not Protected By Attorney-Client Privilege; Federal Court Rules That AI-Generated Documents Are Not Protected by Privilege: Chapman and Cutler LLP; AI Docs Sent By Exec To Attys Not Privileged, Judge Says – Law360; Are AI-generated documents protected from discovery if you send them to your lawyer? One judge says “no” | DLA Piper. ↩︎
2. I could not locate or review the transcript. ↩︎
3. Ben O’Neil is a partner at Quinn Emanuel Urquhart & Sullivan, LLP. ↩︎
4. Presumably, “QE” refers to Quinn Emanuel Urquhart & Sullivan, LLP. ↩︎

Assisted by GAI and LLM Technologies per EDRM’s GAI and LLM Policy.