M&A Discovery In The AI Era: Generative AI Communications And Outputs May Become Litigation Ammunition

The rise of AI use in deal processes, whether for analyzing term sheets, summarizing due diligence findings, or identifying mark-up issues, creates an emerging category of potentially discoverable evidence. No practitioner wants their or their clients' prompts, threads or other content to be used against their clients in earnout litigation, breach claims or other post-closing disputes. The development of jurisprudence concerning AI's implications for the attorney-client privilege and attorney work product doctrines lags behind the dramatic rise of AI's use in connection with deal-making. But courts are beginning to engage with those implications. Below we describe two recent court decisions and offer practical guidance for how deal practitioners may respond in light of the uncertainty they generate.

Background

In decisions issued only one week apart from one another, two judges reached different conclusions regarding whether AI-generated content was protected from disclosure in litigation.

In the case of United States v. Heppner, No. 25-cr-00503-JSR (S.D.N.Y. Feb. 17, 2026), the United States District Court for the Southern District of New York found neither the attorney-client privilege nor the attorney work product doctrine applied. In October 2025, Bradley Heppner, the former CEO and board chairman of Beneficient Company Group—a financial services startup—was indicted on various fraud-related charges. After receiving a grand jury subpoena and retaining defense counsel, but before his arrest in November 2025, he independently used the consumer version of Anthropic's Claude AI platform to generate 31 documents analyzing his legal exposure, potential defense theories and strategic options. Authorities subsequently seized the documents through a search warrant executed at Heppner's residence. The government later moved for a ruling that the documents were not protected from disclosure, and Judge Rakoff granted its motion.

Underlying the Heppner decision regarding the attorney-client privilege are three main points. First, the court held that Claude is not an attorney, and therefore is not capable of establishing an attorney-client relationship with Heppner. Second, the court found that Heppner did not have any reasonable expectation of confidentiality, because Anthropic (i) notifies users that the company collects "inputs" and "outputs" for model training and also (ii) reserves the right to disclose data to third parties including governmental authorities. Third, the court determined that Heppner was not using Claude for the purposes of obtaining legal advice. According to the court, not only did Heppner use Claude "of his own volition" without direction from counsel, but Claude itself disclaims providing legal advice. The opinion noted that when asked whether it can provide legal advice, Claude responded: "I'm not a lawyer and can't provide formal legal advice or recommendations," and recommended that the user consult with a "qualified attorney." As to the attorney work product doctrine, the court concluded that it did not apply because the documents were not prepared by or at the behest of counsel and did not reflect counsel's strategy.

Also just last month, in Warner v. Gilbarco, Inc., No. 2:24-cv-12333-GAD-APP (E.D. Mich. Feb. 10, 2026), the United States District Court for the Eastern District of Michigan concluded that the work product doctrine did protect certain documents generated using AI during litigation. There, a pro se plaintiff brought an employment discrimination case against Gilbarco, Inc. and Vontier Corporation. During discovery, Vontier Corporation moved to compel production of "all documents and information concerning [Warner's] use of third-party AI tools in connection with this lawsuit," arguing that any privilege had been waived by inputting litigation materials into ChatGPT. The court denied the defendant's motion to compel, holding that Warner's AI-generated materials were protected by the work product doctrine and that using ChatGPT did not waive that protection. According to the court, the AI platform is a "tool" and disclosure to it does not result in waiver because it does not materially increase the risk of materials falling into an adversary's hands.

Distinctions between the cases help explain the differing results. Heppner involved an individual that had engaged counsel, but who independently used AI in connection with litigation. That his lawyer was not involved in the preparation of the documents at issue was a significant impediment to either protection attaching. By contrast, Warner involved a pro se litigant who used AI himself to assist in litigation. As a result, Warner did not involve the attorney-client privilege; the plaintiff had no attorney with whom he purported to communicate. And because the Warner plaintiff was pro se, the AI-generated documents revealed the mental impressions of counsel (i.e., himself).

Still, the cases underscore that judges may hold differing views of generative AI tools. Heppner treats them as third parties, disclosure to which fundamentally undermines confidentiality (though that was not the basis for denying work product protection), while Warner treats them as tools (not people), disclosure to which may not so fundamentally undermine confidentiality.

Moving Forward

So where do we go from here? The jurisprudence around generative AI's implications for discovery in litigation will continue to develop and establish more concrete principles. But deal lawyers can't wait for the law to catch up; they must proactively address these issues in connection with deal negotiations and drafting.

After all, Sellers do not want their AI prompts about disclosure obligations to be used in an attempt to establish scienter. By the same token, Buyers do not want due diligence summaries to be used to try and prove "knowledge" that undermines recovery. And neither Sellers nor Buyers want AI-generated projections or analysis to support arguments regarding expected versus actual performance, like a Buyer's analysis of integration synergies that may contradict an earnout target not being achieved.

New contract provisions are worth considering. To provide just a few potential examples:

• Establish Clear AI-Use Deal Policies. Make sure the entire deal team (including outside counsel and other consultants) is clear on your desired AI use protocols, which likely should (i) educate team members about discoverability risks and (ii) distinguish clearly between consumer platforms, on the one hand, and enterprise AI tools with different confidentiality protections on the other.
• Consider AI-Specific Provisions in NDAs and/or Definitive Agreements. We are all aware of the standard privilege assignment/retention clause. Consider various modifications to this standard language, including:
  • Explicitly referencing in deal communications any "AI-Assisted Deal Materials" defined to include: (i) prompts, inquiries or other inputs submitted to any generative AI tool or similar technology, (ii) outputs, documents or other content generated by any generative AI tool or similar technology, and (iii) conversation logs, chat histories or other records of interactions with any generative AI tool or similar technology;
  • An acknowledgement by the parties that, notwithstanding any terms of service, privacy policy or other agreement governing any AI tool, all such AI-Assisted Deal Materials constitute confidential work product prepared in connection with the transactions, which the parties intend to be protected from disclosure;
  • Another acknowledgement that all AI-Assisted Deal Materials created during the relevant period (perhaps from NDA/LOI execution through the closing date) are presumed to have been created at the direction of, or for the purpose of communicating with, legal counsel; and
  • A covenant from the buyer that it will not: (i) directly or indirectly access, review or seek to obtain (including by submitting any subpoena or similar request), (ii) assert waiver of privilege or work product protection in relation to, or (iii) use or introduce into evidence any AI-Assisted Deal Materials.
• Evaluate Pre-Closing Document Removal/Export. Evaluate your document retention policies and, as a seller, determine whether AI-Assisted Deal Materials can be exported and removed from company systems, devices or platform accounts. Mirror long-standing best practices regarding privileged deal communications from company email services and extend them to this new AI context.

The legal landscape around AI and privilege remains unsettled, but the risk is real and present. M&A practitioners should not wait for the courts to more definitively resolve the issue, and should start thinking creatively about how to protect AI communications now.