AI chat history lets AI models recall past interactions. This allows for context-aware conversations and personalized experiences. Users usually are able to manage, export, or delete their chat histories.
But what if an employee uses ChatGPT or another generative AI service in a lawsuit? Can that AI chat histories or other archived personalized AI data discoverable? A recent Michigan case sheds some light on the use of GenAI tools in business operations and employment lawsuits.
Background
On December 6, 2021, the defendants hired the plaintiff Sohyon Warner to a remote-working position, as a Data Analytics Project Manager. The plaintiff, a woman of Korean descent, was replaced on a critical project by a white male without any warning. Around the same time, a black female was also removed from the project. As a result, there were no minorities assigned to the project. The plaintiff complained to Human Resources. At her annual review, she received high marks. The following year, she was again removed from an important project, and she again made a complaint to HR.
On February 3, 2023, the plaintiff filed a Charge of Discrimination with the Equal Employment Opportunity Commission, complaining that she had been discriminated against based on her national origin, race, ethnicity, and gender. That same month, she told her supervisor and Human Resources of her pending Charge of Discrimination. A few weeks later, she was terminated without basis.
A plaintiff filed a lawsuit in September 2024, claiming violations of Title VII, 42 U.S.C. § 2000e et seq., 42 U.S.C. § 1981, and Michigan's Elliott-Larsen Civil Rights Act. She alleged discrimination and retaliation against her former employer, Gilbarco, Inc.
United States Magistrate Judge Anthony P. Patti heard motions that included the defendants' motion to compel and request for sanctions. The defendants sought production of the plaintiff of all documents and information concerning her use of third-party AI tools in connection with this lawsuit. This included all “prompts/queries submitted; all documents, emails, text, or data uploaded or input to any AI tool; all outputs/responses received; all account activity logs, usage histories, and exportable chat histories or archives reflecting any of the foregoing.” The motion also sought to overrule the plaintiff’s attorney–client privilege and work-product objections to the AI materials.
The defendants argued that the plaintiff admitted to extensive use of generative AI tools in this case. They asserted that her written queries “almost certainly” contain admissions and other relevant evidence that concern her allegations in this lawsuit. Nonetheless, in written discovery, the defendants said that the plaintiff declined to produce any materials concerning her use of third-party AI tools to draft filings and otherwise assist her prosecution of this case, and she refused to answer interrogatories concerning her use of ChatGPT. Specifically, the defendants wrote:
During Plaintiff’s deposition, she admitted she used ChatGPT to draft filings; admitted she uploaded emails and other case-related materials to ChatGPT; admitted there are records of her prompts/queries and outputs; and further admitted she deleted some of those records. Production of that information is likely to reveal that Plaintiff uploaded documents to a third-party tool that may not have been produced in litigation but are relevant to the claims and defenses of the parties. Despite those admissions, Plaintiff continues to withhold all such materials and has not provided a privilege log.
Further, the defendants said that the plaintiff admitted to using ChatGPT to draft filings, and that she uploaded case-related emails and other materials many times. Her prompts/queries and the AI outputs are directly relevant to issues that include, at a minimum:
- the authenticity and authorship of filings and assertions made in this case;
- potential admissions against interest embedded in her prompts (for example, she could have identified weaknesses in her claims and position, described those weaknesses, and asked the third-party generative AI tool to address and minimize them, or she could have admitted elements of her prima facie claims away);
- whether she disclosed or publicized information pertaining to the case to a third party; (iv) her preservation of evidence; and
- her credibility.
Further, the defendants argued that the plaintiff could have made statements to the generative AI tool that directly contradict her written statements during her employment or testimony at deposition.
The plaintiff’s deposition confirmed the existence of responsive materials and that they were within her possession, custody, or control via her ChatGPT account history.
The defendants say that the plaintiff offered no valid basis to withhold these records. Her pro se status and claim that these records were protected under the work product doctrine were inapplicable, they contended.
The Court’s Ruling
The information sought is not discoverable, Judge Patti held. Moreover, it wasn’t relevant or even if marginally relevant. Plus, the judge said it wasn’t proportional. The Court said that to the extent that the defendants asked the Court to "overrule the plaintiff's attorney-client privilege and work-product objections to the AI materials," or alternatively, "if the Court sustains any privilege or protection as to any item, require the plaintiff, within seven days, to serve a Rule 26(b)(5)(A)-compliant privilege log," the request was denied.
Judge Patti went on to reason that even if this information were discoverable, it was subject to protection under the work-product doctrine, which the plaintiff was allowed to assert.
Moreover, to the extent the defendants argued that the plaintiff waived the work-product protection by using ChatGPT, the work-product waiver must be a waiver to an adversary or in a way likely to get in an adversary's hand, the judge explained.
“And ChatGPT (and other generative AI programs) are tools, not persons, even if they may have administrators somewhere in the background.”
As the plaintiff noted in her response, the defendants' motion “asks the Court to compel the plaintiff's internal analysis and mental impressions—i.e., her thought process—rather than any existing document or evidence, which is not discoverable as a matter of law. The motion seeks intrusive post-discovery production based on speculation about what might exist in the plaintiff's internal drafting process, untethered from Rule 26 relevance, disregarding the heightened protection afforded to opinion work product, and improperly attempting to manufacture a waiver where none exists. At its core, the defendants' request is a fishing expedition. . . .”
The Court agreed with the plaintiff that the pursuit of this information is “a distraction from the merits of this case,” and that the defendants' theory, which was supported by no case law but only a legal newspaper article posing rhetorical questions, “would nullify work-product protection in nearly every modern drafting environment, a result no court has endorsed.”
“In the end, both sides of this dispute seek to obtain each other's thought processes, while shielding their opponent from discovery of their own,” the judge wrote.
The Court upheld the protections afforded to the thought processes and litigation strategies of both sides and ordered production of neither. Warner v. Gilbarco, Inc.(U.S. Dist. Ct. E.D. Mich. 2/10/2026).
Takeaway
Courts will undoubtedly continue to determine how AI tools fit within legal doctrines of privilege, confidentiality, and waiver.
The district court in Warner found that the ChatGPT data was protected by work-product. Nonetheless, business owners should consider adopting a comprehensive policy concerning AI governance.
For questions regarding the use of AI tools in your business, contact us at Eanet, PC.