A recent order from Magistrate Judge Thomas O. Farrish of the U.S. District Court for the District of Connecticut addresses an issue many companies are only beginning to confront: when an expert witness uses artificial intelligence as part of their analytical workflow, the AI prompts, queries and related materials may be discoverable. In Conservation Law Foundation, Inc. v. Shell Oil Company, the court compelled production of AI prompts and queries used by a testifying expert, treating them as part of the expert’s discoverable methodology under Federal Rule of Civil Procedure 26.
If an expert uses AI to narrow documents, analyze data or otherwise inform opinions, courts are increasingly likely to require transparency into how that AI was used. Failure to preserve or accurately disclose those materials can expose a company to sanctions, including expert preclusion and adverse jury instructions.
The Court’s Ruling in Plain Terms
The dispute arose after the plaintiff’s expert acknowledged using AI tools hosted in a Microsoft Azure environment to filter and identify potentially relevant documents from the defendants’ production. The defendants sought discovery of the prompts and queries used in that AI process, along with related materials showing how the AI was configured and what data it processed.
The plaintiff resisted on three main grounds. First, it argued AI prompts were outside the scope of Rule 26 discovery. Second, it relied on a Rule 29 agreement between the parties limiting discovery of expert “notes, drafts, or communications.” Third, it asserted no additional responsive materials existed because the expert had used only “search terms,” not AI “prompts.”
Judge Farrish rejected each argument in his ruling on May 18, 2026. He emphasized that an expert’s methodology is fair ground for discovery, and that the process used to reduce a large document set into a subset for analysis is part of that methodology. Relying on existing federal authority, the Court treated an AI-assisted document culling no differently from other analytical tools an expert might use.
The court also found the Rule 29 agreement was not “quite clear” enough to shield AI prompts from discovery. Where discovery is otherwise proper under Rule 26, an ambiguity in a discovery limiting agreement will not carry the day. Finally, the court ordered production despite the plaintiff’s claim no prompts existed, noting the expert’s own assistant had referenced “prompts” in a declaration. That inconsistency gave the defendants a concrete reason to doubt the completeness of the plaintiff’s production. The actual text of the Court’s order is here.
Why AI Changes the Expert Discovery Analysis
From a discovery perspective, AI differs in meaningful ways from traditional expert tools. Conventional software, such as spreadsheets or statistical packages, generally produces transparent and repeatable outputs. AI systems, particularly large language models, introduce discretion and variability at the input stage. How a prompt is framed, what context is provided and what parameters are set can materially influence the output.
Courts increasingly view those inputs as the functional equivalent of an expert’s assumptions or analytical steps. If a prompt shapes what documents are identified or what conclusions are suggested, it can be considered part of the “facts or data considered” by the expert. That places it squarely within FRCP 26(a)(2)’s disclosure regime.
AI also raises reproducibility and reliability questions that are central to Daubert and Rule 702 analysis. If an opposing party cannot see what went into the AI system, it cannot meaningfully test whether the expert’s approach was reliable or whether important information was overlooked.
Broader Federal Trends on AI in Litigation
The Conservation Law Foundation order fits within broader federal trends. Courts have been very clear that existing procedural and evidentiary rules apply to AI just as they do to other technologies. Expert disclosures still must explain the “how” and “why” of an opinion, even when AI is involved. Preservation duties extend to AI generated Electronically Stored Information, including prompts, outputs and relevant logs, once litigation is reasonably anticipated.
At the same time, courts have shown heightened sensitivity to confidentiality risks when AI tools are used. Uploading an opponent’s confidential documents to third party AI platforms can raise protective order concerns, but the use of enterprise grade security measures and data handling restrictions can mitigate those concerns.
Finally, recent rulemaking efforts, such as the proposed Federal Rule of Evidence 707 addressing machine generated evidence, signal judges are thinking carefully about how AI outputs should be evaluated for reliability and admissibility. While proposed Rule 707 focuses on evidence offered without a human expert, it underscores a broader judicial expectation: AI derived conclusions must be explainable and defensible.
Practical Precautions for In House Counsel
The lesson from this decision is not to avoid AI, but to manage it deliberately. In house counsel should assume that if a testifying expert uses AI in a meaningful way, that use will be explored in discovery.
At the engagement stage, companies should require experts to disclose whether and how they intend to use AI tools. Expert engagement letters should address AI explicitly, including preservation, verification and compliance with protective orders. Counsel should be clear that AI prompts, outputs and related workflow materials may need to be preserved and produced. Half the battle will be making sure the expert is aware his or her prompts and outputs could be discoverable.
Experts should be prepared to explain their AI use in plain language. Transparency, documentation and verification are the best defenses to challenges based on AI methodology.
A Brief Note on State Courts
Although this decision arises in federal court, state courts are likely to follow similar principles, particularly those states that model their discovery and expert rules on the federal system. While the details will vary by jurisdiction, in house counsel should expect state court judges will also demand clarity around AI assisted expert work and will be receptive to arguments grounded in reliability, methodology and fairness.
Conclusion
The Conservation Law Foundation order is an early but important signal: AI does not insulate expert work from discovery; it can expand it. Courts are applying familiar rules to unfamiliar tools, and while the holding is not surprising to litigators, it very well could surprise corporate clients and experts. For in house counsel, the safest course is being proactive. By setting expectations with experts, preserving AI related materials and ensuring confidentiality safeguards are in place, companies can take advantage of AI’s efficiencies without inviting unnecessary litigation risk.
Our Artificial Intelligence Industry Team tracks AI court decisions and regulations throughout the country. If you have questions or concerns about AI-related matters, please reach out to attorney Brendan M. Palfreyman at (315) 214-2161 and bpalfreyman@harrisbeachmurtha.com, or the Harris Beach Murtha attorney with whom you most frequently work.
This alert is not a substitute for advice of counsel on specific legal issues.
Harris Beach Murtha’s lawyers and consultants practice from offices throughout Connecticut in Bantam, Hartford, New Haven and Stamford; New York State in Albany, Binghamton, Buffalo, Ithaca, New York City, Niagara Falls, Rochester, Saratoga Springs, Syracuse, Long Island and White Plains; as well as in Boston, Massachusetts, and Newark, New Jersey.