Generative AI is becoming a discovery issue in litigation. When lawyers, clients, experts, or others use these tools, their prompts and outputs can become the subject of requests for production. Courts have so far not adopted a separate body of AI-specific discovery rules. Instead, they are applying familiar principles of attorney-client privilege and work-product protection to a new set of facts.
The answer often depends on the details. Who used the tool, the type of tool, whether counsel directed the use, and whether it was part of preparing for litigation can all matter. So can what was entered into the system and whether the resulting material influenced a filing or expert opinion.
By the time a discovery dispute arises, those facts are usually set. Counsel who address AI use early are in a better position to protect privileged communications and work product and to avoid arguments that those protections were lost.
Courts Are Testing Privilege and Work Product in the AI Context
At least seven recent decisions have addressed the discoverability of AI-related materials in litigation. Those decisions highlight how courts are applying familiar rules on privilege and work product to new facts and reaching different results depending on the circumstances.
| Case | Who Used AI | Treatment | Holding |
| Tremblay v. OpenAI Inc., No. 23-cv-03223, 2024 WL 3748003 (N.D. Cal. Aug. 8, 2024). | Plaintiffs used ChatGPT during pre-suit testing of their copyright claims. The prompts were crafted by counsel. | Partially protected. | The court held that the prompts were opinion work product because they were “queries crafted by counsel” that contained counsel’s “mental impressions and opinions about how to interrogate ChatGPT.”
But because plaintiffs had disclosed certain ChatGPT-generated examples in their complaint, the court ordered production of the related prompts and outputs. It denied discovery of other testing materials that plaintiffs had not disclosed.
|
| Warner v. Gilbarco Inc., 820 F. Supp. 3d 629 (E.D. Mich. 2026). | A pro se plaintiff used third-party generative AI tools while preparing her employment-discrimination case. | Protected. | The court held that the requested AI materials were protected work product because they reflected the plaintiff’s internal drafting process, analysis, and her thought process in anticipation of litigation.
The court also rejected the defendants’ waiver argument, explaining that work-product waiver requires disclosure “to an adversary or in a way likely to get in an adversary’s hand,” and that ChatGPT is a tool, not a person, so using it does not put the plaintiff’s mental impressions in an adversary’s hands.
|
| United States v. Heppner, 820 F. Supp. 3d 292 (S.D.N.Y. 2026). | A criminal defendant used Claude on his own after receiving a grand jury subpoena, without counsel’s direction. | Not protected. | The court held that the defendant’s AI communications were not privileged because “[i]n the absence of an attorney-client relationship, the discussion of legal issues between two non-attorneys is not protected by attorney-client privilege,” and were not protected work product because the documents “were prepared by the defendant on his own volition,” meaning he “was not acting as his counsel’s agent when he communicated with Claude.”
|
| Morgan v. V2X Inc., No. 25-cv-01991, 2026 WL 864223 (D. Colo. Mar. 30, 2026). | A pro se plaintiff used an AI tool while preparing his employment discrimination case. | Partially protected. | The court held that Rule 26(b)(3) protects a pro se litigant’s AI use because “a party’s own mental impressions are nevertheless protected,” and that use of AI does not waive protection since “work product protections are typically waived by disclosure to an adversary.”
The court nonetheless ordered disclosure of the AI tool’s name, since the plaintiff had not shown “disclosing the name of an AI tool would reveal [his] mental impressions or case strategy.”
|
| Conservation Law Foundation v. Shell Oil Co., Civil No. 3:21-cv-00933 (D. Conn. May 18, 2026). | Testifying expert used AI to analyze documents and prepare a report. | Discoverable at the magistrate-judge level, but stayed pending district-court review. | The court treated the expert’s AI prompts as part of the disclosed basis and methodology for the expert’s opinions.
|
| Assini v. Hayward, 2026 WL 1677232 (Sup. Ct. Nassau Cnty. June 4, 2026). | A pro se defendant used ChatGPT to help draft filings and prepare his defense in a civil litigation dispute. | Protected. | The court quashed a subpoena to OpenAI, finding Morgan persuasive that a pro se litigant’s “use of AI closely resembles the kind of confidential, strategy-laden iterative work product” the privilege protects, and distinguishing Heppner as a criminal case subject to different discovery rules.
|
| Shealy Jr. v. Seaside Investments LLC, No. 2684CV00799-BLS2, 2026 WL 2122161 (Mass. Super. June 16, 2026). | A represented plaintiff had his romantic partner, uninstructed by counsel, upload litigation documents to ChatGPT and generate a response letter. | Not protected. | The court held the ChatGPT output was not work product because a romantic partner with “no substantive connection to the litigation or to Plaintiffs’ counsel” is not a “representative” under the doctrine.
The court also distinguished Morgan and Warner as involving pro se litigants rather than a represented party’s non-agent intermediary.
It also held the AI output was not opinion work product because “AI is a tool not a person,” and disclosed nothing of the mental impressions, conclusions, opinions, or legal theories of an attorney or other representative.
|
Early Planning Can Preserve Privilege and Work Product
This initial wave of cases carries real consequences. Litigants have been ordered to produce materials they thought were protected, while experts may have their methodologies challenged because of how they used AI.
In each case, the dispute might have been avoided or better managed by considering at the start how AI could factor into the discovery of relevant information. The following practice points can help practitioners address those issues before a dispute arises.
Identify Existing AI Use at the Beginning of the Matter
Early knowledge is key, both concerning how the client uses AI in its ordinary operations and how the client may have used AI in connection with the developing dispute. Key personnel should be interviewed to determine which tools were used, what information was entered, and whether any output made its way into a filing or business decision. By identifying these issues early, counsel can manage these risks before they become facts in the record.
Investigate AI Use by Vendors and Other Third Parties
Clients naturally focus on what their own employees did. Vendors, consultants, and e-discovery providers, however, may have used AI on the same matter without anyone asking.
Hidden third-party AI use carries the same risks as direct party use. It may also be harder to detect and address after the fact, particularly if a vendor’s platform has different retention practices or access controls.
Establish Consistent Rules for Matter-Related AI Use
Practitioners would benefit from upfront conversations with lawyers, experts, and vendors to ensure that AI is used consistently and thoughtfully. Without such a conversation, different individuals may make different choices on the same matter.
That may create inconsistent records and unpredictable exposure that is difficult to explain and harder to fix. Written guidance can help establish what tools may be used, for which purposes, and with what categories of information.
Treat Platform Selection as a Litigation Decision
A public-facing tool that retains prompts and outputs creates different risks from a closed enterprise system. Before AI work begins on a matter, practitioners should understand the provider’s retention practices, confidentiality terms, and access controls. Platform choice may affect whether information is preserved, who can access it, and whether an opposing party can argue that confidentiality or privilege has been compromised.
Set Written Parameters for Expert AI Use
If an expert uses AI to analyze documents or prepare a report, prompts and outputs may become discoverable as part of the disclosed basis and methodology for the expert’s opinions. Conservation Law Foundation illustrates that potential risk.
Defining in writing what tools experts may use and what information they may enter can reduce potential exposure. It may also be appropriate to ask opposing experts similar questions in discovery if AI may have shaped their analysis.
Address AI Specifically in Preservation Plans
A standard litigation hold may not automatically capture AI-related materials. Those materials may live in different places and disappear under different retention schedules.
If AI has been used on a matter, practitioners should identify specifically what needs to be preserved and where it is stored. Vague instructions are unlikely to be enough where prompts, outputs, uploaded files, and platform records may each exist in other locations.
Include AI in Protective Orders and ESI Protocols
Leaving AI out of discovery protocols can invite disputes that might otherwise be avoided. Parties can agree upfront on whether prompts, outputs, platform identities, and uploaded files fall within the scope of discovery.
Protective orders can also address whether confidential material may be entered into AI systems at all. Clear provisions may reduce later disputes over the treatment of sensitive information and the production of AI-related records.
Review Contracts with AI-Using Third Parties
If a vendor or consultant may use AI on client work, practitioners should review what the engagement agreement says about data use, indemnity, and liability. Those provisions may determine the parties’ rights if a third party’s AI use leads to a disclosure or downstream claim.
Contractual protections are particularly important when the client has limited visibility into the third party’s tools. In some circumstances, the contract may be the only available recourse.
Use Targeted Discovery to Investigate the Other Side’s AI Use
In the right case, a focused document request or deposition topic directed at AI use can develop useful facts. Such discovery is most likely to matter when AI may have shaped a pleading, an expert opinion, or the handling of confidential material. The goal should be targeted inquiry rather than open-ended investigation. The relevant questions will depend on how AI was used in the specific matter at hand.
Revisit the Approach as the Law Develops
The cases surveyed here represent an initial wave of decisions. More cases are likely to follow, and some may change the analysis. Protocols and instructions that seem adequate today may need revision after the next decision. Practitioners should therefore monitor this area as the case law develops and revisit matter-specific guidance when circumstances change.
AI Governance Should Be Part of Litigation Strategy
The first wave of AI-discovery decisions offers a straightforward lesson: AI does not displace familiar discovery doctrine, but it can create new factual pathways for discovery disputes. The strongest position is usually built before the tool is used. As courts continue to address these issues, litigators who treat AI governance as part of matter strategy will be better positioned to protect their clients and defend their discovery positions.
