Before You Hit “Enter”: What Recent Privilege Cases Mean for Employers Using AI
Artificial intelligence (AI) is changing the way businesses operate, including the way lawyers and their clients analyze legal issues. But before anyone at your company puts information into an AI system, they should consider whether the information will be protected from disclosure in litigation after they hit “Enter.”
That question carries real consequences for the attorney/client privilege and the work product doctrine. Since February 2026, more than a dozen trial courts have addressed whether a party’s AI prompts, outputs, and uploads are protected from discovery in litigation. While the decisions reached by these courts are highly fact-intensive and, in some cases, inconsistent, they demonstrate that the question of whether sensitive interactions with AI are discoverable in litigation may hinge on who is using the AI, for what purpose, on what platform, and whether legal counsel directed the use.
Examples of Courts Declining Protection for AI-Assisted Materials
Several courts have refused to protect interactions with AI systems from discovery. The leading case on the “no protection” side is United States v. Heppner, 2026 WL 436479 (S.D.N.Y. Feb. 17, 2026). There, a white-collar criminal defendant who was represented by counsel used the consumer version of Claude to analyze information he received from his lawyers and to generate documents relating to his defense strategy. The government sought a ruling from the court that the materials, which had been seized under a search warrant, were not protected by the attorney/client privilege or work product privilege.
The court agreed with the government and found that the materials were not protected by any privilege. In reaching its decision, the court first noted that the documents were not communications between the defendant and an attorney, since Claude was not a lawyer. The court also relied on Claude's publicly available terms and privacy policy, including provisions that permitted Anthropic’s disclosure of user data to third parties, in finding that the defendant had no reasonable expectation of confidentiality in his communications with Claude. Finally, the court held that, while the AI materials were related to the defendant’s legal strategy, they were not created by or at the direction of legal counsel, and thus they did not reveal counsel's legal advice or mental impressions.
The Heppner decision is significant because it illustrates two distinct challenges in protecting exchanges with AI under privilege. First, exchanges with AI may not qualify as attorney/client privileged communications or attorney work product to begin with. However, even if they did, the terms and conditions of a consumer AI platform that allow disclosure of information to third parties may independently defeat any privilege that otherwise could be found to exist.
A Massachusetts court reached a similar result in a civil case in Shealy v. Seaside Investments, LLC, No. 2684CV00799-BLS2 (Mass. Super. Ct., Bus. Litig. Sess. June 16, 2026). There, the plaintiff was involved in a contractual dispute relating to a gas station he sold to the defendants. The plaintiff forwarded the loan documents and a letter from the defendants to his romantic partner, who uploaded them to ChatGPT and used the tool to prepare a response letter and other documents. In discovery, the defendants sought the production of all ChatGPT communications and output. The plaintiff argued that the exchanges with AI were prepared in anticipation of litigation and therefore were protected under the work product doctrine.
The court granted the defendant’s motion to compel discovery and ordered the production of the AI materials. In reaching its decision, the court found that the plaintiff’s romantic partner was not an attorney or other representative whose documents were protected under Massachusetts Rule of Civil Procedure 26(b)(3). The court also relied on the fact that, while the plaintiff had counsel, his lawyers were not involved in the decision to run the documents through ChatGPT. The court held that neither AI queries nor AI output are protected from disclosure under the work product doctrine unless they are performed at the direction of counsel. Finally, the court found that ChatGPT’s output did not constitute opinion work product, stating “AI is a tool not a person. Nothing about what ChatGPT produces disclosed the ‘mental impressions, conclusions, opinions, or legal theories of an attorney or other representative of a party concerning the litigation.’”
The key common thread between Heppner and Shealy is that, in both cases, the party was represented by counsel but chose to use AI independently, without being instructed by counsel to do so. Because the AI-assisted work occurred outside the attorney/client relationship, the exchanges with AI were not privileged.
Examples of Courts Recognizing Work Product Protection
In other cases where a party used AI to prepare for litigation, courts have reached different results and found the information is privileged. For example, in Warner v. Gilbarco, Inc., 820 F.Supp.3d 629 (E.D. Mich. 2026), a self-represented plaintiff in an employment discrimination case used ChatGPT to help draft court filings. The defendants moved to compel production of all materials related to the plaintiff’s AI use. The court denied the motion, holding that the materials were protected work product. In so doing, the court held that work-product waiver requires disclosure “to an adversary or in a way likely to get in an adversary’s hand,” and that using an AI tool did not meet that standard.
The District of Colorado reached a similar result in Morgan v. V2X, Inc., 2026 WL 864223 (D. Colo. Mar. 30, 2026). There, the court found that a pro se plaintiff’s AI communications could reflect mental impressions, opinions, and legal theories protected under Federal Rule of Civil Procedure 26(b)(3). However, the court still required the plaintiff to disclose the name of the AI tool used and prohibited either party from uploading confidential discovery materials into any AI platform that lacks contractual confidentiality safeguards. In reaching that decision, the court recognized that its ruling would practically bar parties from using most, if not all, free or low-cost AI to process confidential information, which would disadvantage pro se litigants. However, the court noted that it “cannot ignore the real risks associated with mainstream tools that persistently collect and store data and could compromise confidentiality.”
A Texas court also extended protection to AI materials in Tate Group Automotive, LLC v. Legacy Automotive Capital, LLC, Case No. 25-BC11B-0020 (Bus. Ct. Tex. June 3, 2026). Unlike the pro se plaintiffs in Warner and Morgan, the AI user in Tate Group was represented by counsel. The defendants argued that work-product protection could not extend to a non-lawyer’s AI interactions, and that uploading legal analysis to ChatGPT waived any privilege protection. The court disagreed, holding that ChatGPT conversations prepared in anticipation of litigation fell within Texas Rule of Civil Procedure 192.5(a)(1), which protects “material prepared or mental impressions developed in anticipation of litigation or for trial by or for a party.” The court also rejected the defendants’ waiver argument, finding that work product is waived only through disclosure to an adversary, or where the risk of access by an adversary is substantially increased. However, while the court protected the AI chat logs from discovery, the court ordered the plaintiff to produce all discovery materials and work product that had been shared with ChatGPT, including materials that had been designated confidential under the protective order, and recommended that the parties amend the protective order to address AI use.
The day after Tate Group, a New York state court reached a similar conclusion in Assini v. Hayward, 2026 NY Slip Op 26086 (Sup. Ct. Nassau County June 4, 2026). There, the plaintiffs subpoenaed OpenAI directly, seeking a pro se defendant’s complete ChatGPT records, including every prompt, upload, and output tied to the case. The court quashed the subpoena, holding that AI-assisted litigation preparation can qualify as protected material under CPLR 3101(d), and expressly declined to follow Heppner. Assini is unique and highly relevant to employers because it involves a demand served not on the company itself, but on the AI provider.
Together, the cases recognizing protection stand for the principle that AI prompts and outputs created because of litigation may qualify as trial preparation materials that are protected from discovery as work product, even if the AI tool is a consumer version with no confidentiality safeguards. However, in light of the contrasting results reached by the courts in Heppner and the other cases where privilege claims were rejected, employers cannot rely on the fact that their litigation-related communications with AI will be protected. Until appellate courts weigh in, there is a significant risk that AI queries and output will be required to be disclosed to adversaries in discovery.
Implications for Employers
The decisions discussed above do not hold that, as a blanket rule, AI use destroys privilege. Instead, they recognize that the outcome may depend on the identity of the AI user, the purpose of the use, the AI platform involved, whether counsel directed the use of AI, and the applicable rules of procedure.
An in-house lawyer who uses an approved enterprise AI platform to assist with legal research or drafting is in a very different position from an HR director who independently pastes a privileged memorandum from outside counsel into a consumer chatbot. The first use is likely privileged; the second use very well may not be, and privilege over the attorney’s memorandum may be waived.
Also, as courts have recognized, consumer-grade AI products and enterprise platforms often have materially different terms governing data retention, model training, access, and disclosure. Those differences can affect whether a court finds that a user had a reasonable expectation of confidentiality. However, that a company may use an enterprise platform does not, by itself, ensure that AI materials will be protected. The company still should consider who is using the system and for what purpose, what information is being provided, and whether counsel directed the work.
The above cases also highlight a separate risk that is unrelated to the privilege issue. HR employees who use AI to evaluate a complaint, summarize an investigation, rank employees for a reduction in force (RIF), or help formulate the rationale for a termination may create records that become relevant in litigation. Those materials may be discoverable, not because privilege was waived, but because it is evidence of the decision-making process, including the employer’s stated reasons for an adverse employment action. Employers should be cautious in using AI for these purposes and should discuss the implications of doing so with counsel.
Practical Recommendations for Employers
Adopt and implement an AI acceptable use policy. Companies should work with counsel to develop an AI acceptable use policy to decrease the risk that privileged or sensitive AI materials will be produced in litigation. Employees should be trained on the use of the policy, including regarding which AI platforms the company has authorized for use, the circumstances when AI may be used, and the types of company information that may be entered into AI systems.
Route AI use on legal matters through counsel. Employees should not enter privileged communications into AI tools without guidance from the legal department. Where AI is used to support investigations or anticipated litigation, counsel should direct that use and document the direction.
Keep litigation use separate from general business use. AI prompts and output generated by AI systems are more likely to be protected when they are created for use in, or in anticipation of, specific litigation matters. Companies should use separate threads for case-specific queries and general business queries in order to avoid disputes over line-drawing.
Thoroughly vet AI vendors and contract terms. Companies should know how an AI provider handles prompts and uploaded documents, including retention, training, access, and disclosure provisions.
Update litigation hold procedures. Litigation hold notices should address AI-generated content. If an employee uses an AI tool to analyze a workplace complaint or draft talking points for a termination meeting, those prompts and outputs may be discoverable, and they should be preserved in the same manner as email and text messages. Companies should also consider whether AI-provider retention policies align with their own preservation obligations, and whether they can retrieve AI data from their providers if a legal hold is triggered.
What Does the Future Hold?
The privilege and discovery issues associated with AI materials are evolving nearly as rapidly as AI technology itself. The decisions discussed above are trial court decisions which provide useful guidance to employers, but which cannot be relied on as legal precedent. As decisions such as these are ruled on by state appellate courts, federal circuit courts, and the United States Supreme Court, the picture for employers should become much clearer. However, in the meantime, employers should work with counsel and take the steps outlined above to best protect their privileged and sensitive information from disclosure.
About Maynard Nexsen
Maynard® is a nationally ranked, full-service law firm with more than 600 attorneys nationwide, representing public and private clients across diverse industries. The firm fosters entrepreneurial growth and delivers innovative, high-quality legal solutions to support client success.