Artificial Intelligence, Privilege, and Work Product: Emerging Risks in the Life Sciences Industry

Introduction

As artificial intelligence (AI) use has become prevalent in nearly every stage of litigation, including pre-litigation efforts, courts and litigants alike have encountered challenges applying longstanding doctrine to new technology. Recent decisions indicate that courts disagree on whether communications with generative AI tools are more like disclosures to a third party or more like the use of traditional word-processing tools. In particular, courts are grappling with whether disclosure of sensitive information to publicly available AI platforms constitutes disclosure to a “third party” sufficient to waive attorney-client privilege or undermine attorney work-product protection.

These developments are especially significant for life sciences companies, which routinely handle highly confidential and proprietary information, including intellectual property, trade secrets, clinical research, regulatory strategy, and commercially sensitive data. Because legal advice in the life sciences sector is often deeply intertwined with technical and scientific information, the use of AI tools in connection with legal and business decision-making presents heightened discovery and confidentiality risks.

Recent decisions from courts across the country show courts approaching these questions differently, with results that often track the particular protection at issue, attorney-client privilege, or work product.

United States v. Heppner: AI Communications and the Limits of Attorney-Client Privilege

As we have discussed, in United States v. Heppner, 820 F. Supp. 3d 292 (S.D.N.Y. 2026), Judge Rakoff addressed whether a criminal defendant’s use of a generative AI platform to develop defense strategies could be protected by either the attorney-client privilege or the work-product doctrine.

After learning he was the target of a criminal investigation, Heppner used the publicly available version of generative AI platform Claude to prepare reports that outlined a potential defense strategy. The government later seized those materials while executing a search warrant. Through (human, retained) counsel, Heppner argued that the materials were privileged because Heppner (1) inputted into Claude information that he had learned from counsel, among other things; (2) created the AI documents for the purpose of speaking with counsel to obtain legal advice; and (3) shared the contents of the AI documents with counsel.

The court rejected Heppner’s attorney-client privilege arguments, concluding that the AI documents “lack at least two, if not all three, elements of the attorney-client privilege.” Id. at 296. The court emphasized that communications with the publicly available version of Claude are not communications with an attorney, are not confidential, and were not made for the purpose of obtaining legal advice from a lawyer. Although the last element was a “closer call” because, as defense counsel argued, Heppner communicated with Claude for the “express purpose of talking to counsel,” defense counsel nonetheless conceded that these communications were not “at the suggestion or direction of counsel.” Id. at 297. Had that been the case, the court noted that Claude “might arguably be said to have functioned in a manner akin to a highly trained professional who may act as a lawyer’s agent within the protection of the attorney-client privilege.” Id. However, the communications were made of Heppner’s own volition, and Claude itself “disclaims providing legal advice.” Id. Accordingly, the communications were not privileged at the time they took place and could not “somehow [be] alchemically changed into privileged ones upon being shared with counsel.” Id. Moreover, the court held that even if the communications that Heppner provided to Claude were privileged, Heppner waived any such privilege by sharing the information with the publicly available version of Claude and Anthropic “just as if he had shared it with any other third party.” Id. at 297 n.3. Indeed, the court noted that Claude and Anthropic’s “written privacy policy to which users of Claude consent provides that Anthropic collects data on both users’ ‘inputs’ and Claude’s ‘outputs,’ that it uses such data to ‘train’ Claude, and that Anthropic reserves the right to disclose such data to a host of ‘third parties,’ including ‘governmental regulatory authorities.’” Id. at 296. Therefore, Heppner lacked a reasonable expectation that his inputs would not be shared with other third parties. Id. at 297 n.3.

The court further concluded that the work-product doctrine did not apply because, even if the AI reports had been prepared in anticipation of litigation, they were not prepared at the behest of counsel and did not reflect defense counsel’s strategy. Id. at 298.

Heppner reflects a strict, fact-bound application of the attorney-client privilege to communications with publicly available generative AI. If adopted more broadly, this reasoning could have substantial implications for clients who use publicly available AI tools in connection with legal matters, even where external legal counsel has been retained or in-house legal counsel has been involved in a brewing dispute. In particular, company employees need to be cautious not to conduct ostensibly legal research using publicly available AI tools without the direction or instruction of counsel. Such work is unlikely to be privileged in the first place because it does not involve communication with an attorney, and to the extent an employee inputs advice already received from counsel, that disclosure risks waiving the privilege the advice would otherwise enjoy.

AI and The Attorney Work Product Protection

While Heppner, one of the first in-depth judicial opinions addressing the interplay between the use of AI and the attorney-client privilege, applied those protections strictly, other courts have taken relatively expansive views of the attorney work product doctrine and AI-assisted litigation preparation. In both Warner v. Gilbarco, Inc., 820 F. Supp. 3d 629 (E.D. Mich. 2026), and Morgan v. V2X, Inc., No. 25-CV-01991-SKC-MDB, 2026 WL 864223 (D. Colo. Mar. 30, 2026), the courts concluded that pro se plaintiffs could invoke the work-product doctrine with respect to materials generated through the use of publicly available AI tools.

Although aspects of the analyses in Morgan and Warner may be unique to pro se litigants and/or reflective of how one may waive the attorney-client privilege versus the attorney work product privilege, the courts’ expansive treatment of the parties’ expectation of privacy when interacting with AI is notable. In Heppner, the court stated that “in light of Anthropic’s privacy policy,” the defendant “had no reasonable expectation that the inputs would not be shared with other third parties.” Heppner, 820 F. Supp. 3d at 297 n.3. By contrast, in Morgan, the court stated it was “entirely reasonable for a person to expect some privacy and confidentiality when interacting with these tools, even though they understand a third party is behind the tool collecting and storing their information.” 2026 WL 864223, at *5. Similarly, in Warner, the court noted that “the work-product waiver has to be a waiver to an adversary or in a way likely to get into an adversary’s hand … And ChatGPT (and other generative AI programs) are tools, not persons, even if they may have administrators somewhere in the background.” 820 F. Supp. 3d at 636-37. As such, Morgan and Warner reflect a pragmatic understanding of modern technology use. For example, as the court in Morgan questioned: “[t]oday, nearly all electronic interaction passes through third-party systems …. Does that mean that anyone with a Gmail account has forfeited all rights to confidentiality and privacy?” 2026 WL 864223, at *4. This comparison suggests that some courts may be more sympathetic to litigants seeking work product-protection, depending on the factual circumstances, in light of the ubiquity of AI tools.

Open Questions and Unresolved Issues

Prior decisions have addressed whether an attorney’s use of AI tools in connection with legal strategy or litigation preparation qualifies as work product.¹ Several questions remain open, however. Courts have not resolved whether enterprise-grade AI systems, with contractual confidentiality protections and restricted data retention policies, alter the privilege analysis. The pro se litigant cases do not address whether attorneys will receive the same work-product protection if they use publicly available AI tools when creating work product. Nor do they resolve the question that matters most for corporate clients: not whether an employee’s prompt is itself privileged, which Heppner effectively answered in the negative, but whether an employee who inputs a lawyer’s privileged advice into a closed enterprise system waives the privilege, or whether that system’s contractual confidentiality and retention terms support a reasonable expectation of confidentiality that avoids waiver. In Morgan, for example, the court entered a protective order requiring that confidential information be processed only using enterprise-tier AI accounts, which suggests that such accounts carry a more protectable privacy posture than consumer tools. 2026 WL 864223, at *7. But the court did not address privilege, and the question remains unresolved.

Accordingly, companies and attorneys alike must be cautious when using AI to conduct pre-litigation case assessments or for daily use. For example, if an employee uses a publicly available AI tool to summarize or take notes of a meeting at which counsel provided legal advice, inputting that advice into the tool could be treated as a third-party disclosure that waives a privilege the communication would otherwise enjoy. As courts continue to confront these issues, privilege analyses may increasingly turn on the contractual and privacy terms governing the particular AI use (including whether the platform was used as part of an enterprise license versus publicly available), the platform’s data retention and training practices, the nature of the information disclosed, and whether counsel directed or supervised the AI-assisted work.

Practical Implications for Life Sciences Companies

The implications of these developments are particularly acute for life sciences companies. Life sciences companies have widely adopted AI tools, and relevant legal advice frequently involves proprietary scientific information, regulatory strategy, intellectual property, clinical trial data, and commercially sensitive research that may be subject to discovery in fact-intensive post-acquisition litigation relying on expert opinions, including milestone and earnout disputes.

Companies should assume that information shared with publicly available generative AI tools will be subject to discovery in a future litigation and that such interactions could constitute disclosure to a third party sufficient to waive privilege or work product protections. For example, in Fortis Advisors, LLC v. Krafton, Inc., the Delaware Court of Chancery ruled against Krafton, finding that it breached an acquisition agreement that provided for contingent earnout payments and allowed the founders and CEO to be terminated only for cause. 354 A.3d 906, 953 (Del. Ch. 2026). Krafton’s CEO had used ChatGPT for strategic advice to avoid earnout payments and relied on that advice. Id. at 927-28. In ruling against Krafton, the court rejected Krafton’s explanation for firing certain founders, basing its opinion, in part, on evidence of the CEO’s ChatGPT-designed strategy to force a deal on a term of the agreement or execute a takeover. Id. at 941 n.336. Accordingly, the court stated it would not permit Krafton to “use the after-acquired evidence doctrine to fabricate cause where the evidence shows the termination decision was made for different reasons.” Id. at 942.

The Fortis decision underscores the practical realities of AI usage: AI interactions may be highly probative evidence concerning corporate intent, strategy, and decision-making. Separate from any privilege or waiver question, AI-generated content may create discoverable information that plaintiffs or regulators later use to challenge a company’s stated rationale for business decisions. Prompts may even constitute discoverable expert reliance materials under the Federal Rules. Courts have also restricted how AI tools may be used in litigation. In Jefferies v. Harcros Chemicals, Inc., the court barred parties from uploading any discovery materials to publicly available AI tools, citing the inability to claw back information once it is incorporated into the model. Nos. 25-2352-KHV-ADM, 25-2569-KHV-ADM (D. Kan. 2026), ECF No. 152 at 7.

Best Practices and Risk Mitigation

Given the unsettled legal landscape, companies should always exercise caution when using generative AI tools in connection with legal or business matters, and counsel should adopt clear governance measures regarding AI use, particularly in legal contexts.

Key considerations include:

• Avoid inputting privileged communications, litigation strategy, trade secrets, or confidential technical information into publicly available AI platforms.
• Use enterprise tools governed by contractual confidentiality, no-training, and limited-retention terms.
• Implement internal policies and trainings governing employee and in-house legal use of generative AI, including regarding privilege and confidentiality risks associated with AI platforms, and ensure close coordination between the business and legal departments on these issues.
• Ensure that legal advice is clearly segregated from business discussions when using AI-assisted workflows.
• Consider the role of outside counsel in supervising or directing AI-assisted internal investigations and litigation preparation.
• Carefully evaluate vendor terms governing data retention, training, and access rights.