Source: JD Supra

A recent federal order in Conservation Law Foundation, Inc. v. Shell Oil Company, issued by Magistrate Judge Thomas O. Farrish of the U.S. District Court for the District of Connecticut, compelled production of AI prompts and queries used by a testifying expert, treating those materials as part of the expert’s discoverable methodology under Federal Rule of Civil Procedure 26. The dispute centered on the expert’s use of AI tools hosted in a Microsoft Azure environment to filter and identify potentially relevant documents from defendants’ production; defendants sought the prompts, configuration details and related processing data used in that AI-assisted culling. The plaintiff resisted on three grounds — that prompts were outside Rule 26, that a Rule 29 agreement limited disclosure of expert notes/drafts/communications, and that only “search terms,” not prompts, had been used — but Judge Farrish rejected each argument in his May 18, 2026 ruling.

The court treated AI-assisted document reduction like any other analytical tool, finding that prompts can function as the expert’s assumptions or analytical steps and therefore fall within FRCP 26(a)(2)’s disclosure regime. The judge also held the Rule 29 agreement was not sufficiently clear to bar discovery where Rule 26 otherwise permitted it, and ordered production after noting an assistant’s declaration referenced “prompts,” undermining the plaintiff’s claim of no responsive materials. The opinion highlights consequences for failure to preserve or disclose AI workflow materials, including potential sanctions such as expert preclusion and adverse jury instructions, emphasizes reproducibility and Daubert/Rule 702 concerns when AI inputs are hidden, notes preservation duties for AI-generated ESI (prompts, outputs, logs) once litigation is anticipated, and acknowledges confidentiality risks from uploading opposing parties’ documents to third-party AI platforms while pointing to enterprise security measures and data-handling restrictions as possible mitigations. The piece also references broader federal trends and recent rulemaking efforts (for example, proposed Federal Rule of Evidence 707) reflecting judicial attention to machine-generated evidence and the expectation that AI-derived conclusions be explainable and defensible.