By Phil Favro, Contributing Author for HaystackID
Generative artificial intelligence and long-standing evidentiary protections have recently collided, as evidenced by two new cases: United States v. Heppner and Warner v. Gilbarco. Heppner concluded that neither the lawyer-client privilege nor the attorney work product doctrine applied in a criminal matter to AI content generated in response to a defendant’s inputs that were not made at the direction of his lawyers. In contrast, Warner determined that work product applied to AI-created materials for a pro se plaintiff and forbade the discovery of this information in the context of an employment discrimination lawsuit.
While Heppner and Warner reached opposite results based on their unique factual and legal contexts, they provide insights on how clients and counsel may consider approaching the discovery process when using AI tools. This is particularly the case when clients wish to use AI to search for, identify, and review electronically stored information (ESI) to satisfy production obligations in discovery. Collectively, these cases—particularly Heppner—spotlight the need for counsel to adopt protocols that ensure they ultimately drive the discovery process on behalf of the client. As part of that process, counsel should consider the role of service providers in effectuating the use of AI tools for document review and determine how they will oversee and supervise providers’ use of those tools.
Heppner Rejects Privilege and Work Product Claims
Heppner is a criminal proceeding in which the government charged the defendant with securities fraud and wire fraud.[1] In connection with Heppner’s arrest, FBI agents seized electronic devices and accounts from the defendant’s home, including 31 documents generated by Claude, the AI platform that Anthropic operates.
The defendant claimed attorney-client privilege and work product over the 31 documents, asserting they were created after the grand jury convened, but before it issued its indictment and the defendant’s arrest. The defendant maintained that he generated the documents in anticipation of the indictment and to develop a defense strategy. In particular, the defendant argued that the information he shared with Claude included details obtained from his lawyers. In addition, the documents at issue were generated so the defendant could receive legal advice, and he accordingly provided the 31 documents to his counsel. The government disagreed, arguing that neither immunity applied and that the defendant should produce the 31 withheld documents.
The court sided with the government, holding that neither the privilege nor work product applied to the 31 AI-generated documents. In ruling against the defendant, the Honorable Jed Rakoff, District Judge for the U.S. District Court, the Southern District of New York, characterized the issue as one of “first impression nationwide” and framed the dispute as follows: “whether, when a user communicates with a publicly available AI platform in connection with a pending criminal investigation, are the AI user’s communications protected by attorney-client privilege or the work product doctrine?” The court answered the question succinctly: no.
Defendant’s Prompts Do Not Meet Privilege Requirements
Regarding the issue of privilege, the court discussed the three required elements to establish a claim of privilege: communications that (1) transpired between the client and retained counsel; [2] were intended to be and were in fact confidential; and (3) were made for the purpose of obtaining or rendering legal advice. Judge Rakoff found that the defendant’s claim failed to meet any of these elements.
First, the court determined that the privilege did not apply because the defendant’s communications were with Claude and not with the defendant’s lawyers. In so doing, Judge Rakoff acknowledged the position taken by some commentators that Claude and other AI platforms are analogous to cloud-based word processing platforms and that their non-attorney status is not an issue. Nevertheless, the court dismissed this analogy, finding it undermined the application of privilege. Because the privilege only applies to relationships involving “a licensed professional who owes fiduciary duties and is subject to discipline,” the computer software analogy is misplaced because “[no] such relationship exists, or could exist, between an AI user and a platform such as Claude.”
Next, the court found that the defendant did not establish that his communications with Claude were confidential because Anthropic’s privacy policy clearly disclaimed confidentiality. Among other things, the policy indicated that prompts and outputs could be disclosed to third parties for both AI model training and—“even in the absence of a subpoena”—legal matters. Given that Claude retains and shares inputs and outputs with third parties “in the normal course of its business,” the court reasoned that the defendant could have no reasonable expectation of confidentiality in his communications with Claude. Moreover, Judge Rakoff explained that the 31 documents were not analogous to a client’s “confidential notes” shared with a lawyer, as the defendant first communicated those notes as prompts to Claude, a third party.
Finally, the court found the claim failed to meet the third requirement because the defendant could not show that he sought to obtain legal advice from Claude. Judge Rakoff suggested his decision on this issue was a “closer call.” Had the defendant communicated with Claude on the advice of counsel, the outcome perhaps could have been different. Pursuant to the Second Circuit’s line of authority under United States v. Kovel, 296 F.2d 918 (2d Cir. 1961), Judge Rakoff suggested that the AI platform could have operated by analogy like “a highly trained professional who may act as a lawyer’s agent within the protection of the attorney-client privilege.” Nevertheless, that circumstance was not present as the defendant communicated with Claude of his own volition and not at the direction of his counsel.
Defendant’s Prompts Are Not Work Product
The court was equally dismissive of the defendant’s work product claim. Even assuming the 31 documents were prepared in anticipation of litigation, the court found they were not prepared “at the behest of counsel.” Nor did the documents memorialize “defense counsel’s strategy.” In response to the defendant’s assertion that work product protected non-lawyer generated materials from discovery, the court expressed its concern that those materials fell outside the purpose of the work product immunity: “While it is true the work product doctrine may apply to materials generated by non-lawyers, the Second Circuit has repeatedly stressed that the purpose of the doctrine is to protect lawyers’ mental processes.” Because the documents did not reflect counsel’s mental processes or strategy at the time of creation, they did not fall within the policy goal of safeguarding a lawyer’s “zone of privacy” in which to work, free from the interference of adversaries.
Warner Sustains Plaintiff’s Work Product Claim
In contrast to Heppner, Warner resulted in a more favorable outcome for the AI user, though in a distinct procedural and factual context.[2] In this employment case, the defendant filed a motion to compel written responses regarding a pro se plaintiff’s use of “third-party AI tools” (including ChatGPT) in this lawsuit. In response, the court denied the motion, concluding the request was untimely and sought information that was work product, lacked relevance, and was not proportional to the needs of the case.
In particular, U.S. Magistrate Judge Anthony Patti determined that the plaintiff had a right to assert work product protection over materials prepared for litigation. In response, the defendant argued that the plaintiff waived any work product protection by using ChatGPT. Judge Patti disagreed, reasoning that a waiver of work product must be made “to an adversary or in a way likely to get in an adversary’s hands.” As ChatGPT is a tool and not a person, the court found that the defendant’s waiver argument was inapplicable.
Judge Patti ultimately concluded that the plaintiff’s internal thought processes, which were arguably “reformatted” by the software, were “opinion work product” and should not be subject to discovery. Were the court to hold otherwise, Judge Patti opined (without further explanation) that his ruling “would nullify work-product protection in nearly every modern drafting environment.”
The Impact of Heppner and Warner on Privilege, Work Product, and Discovery
Heppner and Warner offer contrasting views on how courts should handle the disposition of privilege and work product claims over AI platform prompts and outputs. Heppner counsels clients and lawyers to exercise caution when using AI platforms whose policies allow those platforms to both retain and share user inputs and AI outputs with third parties. Under those circumstances, Heppner appears to militate against privilege or work product claims over prompts and outputs. In contrast, Warner seems to suggest otherwise, sustaining the plaintiff’s work product claim arising from her use of ChatGPT.
Is Heppner more instructive than Warner on discovery issues?
Heppner is probably more instructive on discovery issues, particularly for clients who wish to use AI to search for, identify, and review ESI. Parties could rely on Warner to support their position that AI platforms like ChatGPT are acceptable for handling the ESI search and identification process. Nevertheless, given the restrictions in standard protective orders on the dissemination of confidential information and the reality that AI platforms with privacy policies like those of Anthropic may disclose user prompts to third parties, the safer course is to follow Heppner and avoid divulging confidential information to AI tools like Claude.
Could Heppner Help Shape AI Document Review Processes?
Even though Heppner did not address civil discovery issues, its rationale—premised on the notion that lawyers should be inextricably intertwined with the client’s use of AI to better merit privilege and work product protection—is applicable to the ESI search, identification, and review process.
For example, clients should ensure that lawyers are driving the process of searching, identifying, and reviewing ESI for responsiveness and privilege in AI document review workflows. Like traditional document reviews, service provider personnel will be involved in this process. However, their purpose is to facilitate, not dictate, strategic decisions on litigation and discovery issues reserved for clients and counsel.
With counsel at the helm, courts—following Heppner—may defer to the client’s established process for AI document reviews. If courts scrutinize the process, though, how things actually play out—where there is often give and take between counsel and service providers on a day-to-day basis—may vary from one case to the next.[3] In light of Heppner, consider the following queries a court may pose about an AI document review process:
- What roles can service provider representatives occupy without impinging on counsel’s role?
- Pursuant to Kovel, may service provider representatives (who are skilled in developing AI prompts) propose inputs to facilitate counsel’s provision of legal advice?
- Should lawyers alone be responsible for developing prompts?
These questions will remain unanswered until squarely addressed by a court. In the meantime and to ensure they are prepared to address downstream privilege and work product issues, lawyers should consider carefully documenting a client’s AI document review process. This includes detailing the role of service providers and their personnel and then describing counsel’s approach for both overseeing and supervising provider involvement. Documenting (and then following) such a top-down process should help substantiate client privilege claims and lawyer assertions of work product.
[1] United States v. Heppner, — F. Supp. 3d —, 2026 WL 436479 (S.D.N.Y. Feb. 17, 2026).
[2] Warner v. Gilbarco, Inc., — F. Supp. 3d —, 2026 WL 373043 (E.D. Mich. Feb. 10, 2026).
[3] See Winfield v. City of New York, , No. 15-CV-05236 (LTS) (KHP), 2017 WL 5664852 (S.D.N.Y. Nov. 27, 2017) (finding—after in camera review—that the defendant’s predictive coding process and training “protected by the work product privilege and, accordingly, is not subject to disclosure”).
About Phil Favro
Phil Favro is the founder of Favro Law PLLC, where he counsels clients on ESI, AI, and discovery issues and serves as a special master, mediator, and expert witness. Phil is nationally recognized for his expertise on ESI, discovery, and information governance, with courts acknowledging his credentials. See, e.g., Oakley v. MSG Networks, Inc., No. 17-CV-6903 (RJS), 2025 WL 2061665 (S.D.N.Y. July 23, 2025). This background makes Phil particularly well-suited to counsel clients and advise courts on information-related issues. As a special master, Phil is acclaimed for his collaborative approach, working with parties to find stipulated solutions to complex issues. For disputes that require adjudication, he is renowned for the clarity and vigor of his written dispositions, which are available on legal search engines.
About HaystackID®
HaystackID® solves complex data challenges related to legal, compliance, regulatory, and cyber requirements. Core offerings include Global Advisory, Cybersecurity, Core Intelligence AI™, and ReviewRight® Global Managed Review, supported by its unified CoreFlex™ service interface and eDiscovery AI™ technology. Recognized globally by industry leaders, including Chambers, Gartner, IDC, and Legaltech News, HaystackID helps corporations and legal practices manage data gravity, where information demands action, and workflow gravity, where critical requirements demand coordinated expertise, delivering innovative solutions with a continual focus on security, privacy, and integrity. Learn more at HaystackID.com.
Assisted by GAI and LLM technologies.
SOURCE: HaystackID
