AI Regulation Tracker / Litigation

A Federal Court Says an Expert's AI Prompts Are Discoverable, Then Stays Its Own Order

Regulatory summary: On May 18, 2026, Magistrate Judge Thomas O. Farrish of the United States District Court for the District of Connecticut ordered the plaintiff in Conservation Law Foundation, Inc.

In Conservation Law Foundation v. Shell Oil, a Connecticut magistrate judge held that the generative-AI prompts an expert used to cull a document set are part of her methodology and discoverable under Rule 26. The district court then stayed the order while the plaintiff's objection is decided, so the question is live.

Primary source

By Anthony Guerriero, Founder, The Leveraged Years · Reviewed by The Leveraged Years Editorial Desk · Published July 9, 2026 · Last updated July 9, 2026

A Federal Court Says an Expert's AI Prompts Are Discoverable, Then Stays Its Own Order regulation briefing

The Leveraged Years AI Regulation Tracker

Key takeaways

A federal court has now treated the generative-AI prompts behind an expert's document culling as part of the expert's methodology, and therefore as discoverable, rather than as protected work product or informal notes. The expert reportedly used OpenAI's GPT-4o, accessed through a private Microsoft Azure environment, to filter documents Shell had produced into a smaller relevant set. That AI step became the focus of a discovery fight, and the court decided the inputs to it are fair game.
Litigators and expert witnesses in federal cases; e-discovery and litigation-support teams that deploy generative AI for document review; in-house legal and compliance leaders overseeing outside counsel's AI use; and any organization whose experts run large document sets through AI tools. It also matters to AI vendors selling review and analysis tools into the legal market, because prompt-level auditability is now a live litigation issue.
Status: Order issued May 18, 2026 by Magistrate Judge Farrish in the District of Connecticut.
Before an expert uses generative AI on a document set, map and record the workflow: the tool and version, the environment, how documents were loaded, and how prompts were used to filter or analyze. Decide in writing, with opposing counsel where appropriate, how prompts will be treated, and preserve prompt logs so you can produce or defend them if a court follows this reasoning.

Regulatory Briefing

Bindingness: Trial-court discovery order, stayed and under review. This is a ruling by a magistrate judge on a discovery dispute, not an appellate precedent and not a statute or rule change. It binds the parties in this case, subject to the pending objection. Its wider value is persuasive: it is one of the first decisions to address whether an expert's AI prompts are discoverable, so other courts and litigants will look to its reasoning even though it is not controlling authority.
Scope: United States federal civil litigation. The decision applies Federal Rule of Civil Procedure 26 in a single case in the District of Connecticut, within the Second Circuit. It speaks directly to expert discovery and AI-assisted document review, and its reasoning could carry into any federal matter where an expert or party uses generative AI to process evidence.
What Changed: A federal court has now treated the generative-AI prompts behind an expert's document culling as part of the expert's methodology, and therefore as discoverable, rather than as protected work product or informal notes. The expert reportedly used OpenAI's GPT-4o, accessed through a private Microsoft Azure environment, to filter documents Shell had produced into a smaller relevant set. That AI step became the focus of a discovery fight, and the court decided the inputs to it are fair game.
Who This Affects: Litigators and expert witnesses in federal cases; e-discovery and litigation-support teams that deploy generative AI for document review; in-house legal and compliance leaders overseeing outside counsel's AI use; and any organization whose experts run large document sets through AI tools. It also matters to AI vendors selling review and analysis tools into the legal market, because prompt-level auditability is now a live litigation issue.
Current Status: Order issued May 18, 2026 by Magistrate Judge Farrish in the District of Connecticut. The plaintiff filed an emergency motion to stay and a Rule 72(a) objection, and the district court stayed the order pending resolution of that objection. As of the news peg on July 7, 2026, the objection is pending before District Judge Vernon Oliver and the order is not in effect.
Effective Date: Not applicable in the sense of a rule taking force. The operative event is the May 18, 2026 order, whose effect is currently suspended by the district court's stay until the Rule 72(a) objection is decided.
Primary Source: https://www.courtlistener.com/docket/60042265/conservation-law-foundation-inc-v-shell-oil-company/
Secondary Analysis: Mayer Brown and Arnold & Porter published detailed client analyses of the order; Eugene Volokh covered it at Reason's Volokh Conspiracy (May 26, 2026); E&E News by Politico reported on what the case signals for how courts treat AI. Minnesota Lawyer covered the dispute around July 7, 2026.
Workflow Impact: Legal teams should treat AI prompts, prompt logs, and the surrounding workflow as potentially discoverable artifacts. Expert engagements that involve AI-assisted review will need a documented, defensible process, and any claim that prompts are protected notes or communications should be grounded in an express, clearly worded agreement rather than assumed.
Operational Next Step: Before an expert uses generative AI on a document set, map and record the workflow: the tool and version, the environment, how documents were loaded, and how prompts were used to filter or analyze. Decide in writing, with opposing counsel where appropriate, how prompts will be treated, and preserve prompt logs so you can produce or defend them if a court follows this reasoning.

Date	Jurisdiction	Rule	Affected professionals	Status or effective date
2026-07-09	United States (federal, D. Conn.)	A federal court has now treated the generative-AI prompts behind an expert's document culling as part of the expert's methodology, and therefore as discoverable, rather than as protected work product or informal notes. The expert reportedly used OpenAI's GPT-4o, accessed through a private Microsoft Azure environment, to filter documents Shell had produced into a smaller relevant set. That AI step became the focus of a discovery fight, and the court decided the inputs to it are fair game.	Litigators and expert witnesses in federal cases; e-discovery and litigation-support teams that deploy generative AI for document review; in-house legal and compliance leaders overseeing outside counsel's AI use; and any organization whose experts run large document sets through AI tools. It also matters to AI vendors selling review and analysis tools into the legal market, because prompt-level auditability is now a live litigation issue.	Order issued May 18, 2026 by Magistrate Judge Farrish in the District of Connecticut. The plaintiff filed an emergency motion to stay and a Rule 72(a) objection, and the district court stayed the order pending resolution of that objection. As of the news peg on July 7, 2026, the objection is pending before District Judge Vernon Oliver and the order is not in effect.

Source File

Instrument: Order compelling production of expert's generative-AI prompts (discovery ruling under Fed. R. Civ. P. 26)
Authority: United States District Court for the District of Connecticut, Magistrate Judge Thomas O. Farrish
Date: Order issued May 18, 2026; stayed by the district court pending plaintiff's Rule 72(a) objection (filed June 2, 2026); news peg July 7, 2026
Citation: Conservation Law Foundation, Inc. v. Shell Oil Co., et al., No. 3:21-cv-00933 (D. Conn. May 18, 2026); order cited as ECF No. 970 per Mayer Brown analysis
Primary Source: CourtListener docket, https://www.courtlistener.com/docket/60042265/conservation-law-foundation-inc-v-shell-oil-company/ ; govinfo case package, https://www.govinfo.gov/app/details/USCOURTS-ctd-3_21-cv-00933
Corroboration: Mayer Brown, Court Orders Disclosure of Expert Witness's AI Prompts: What Litigators Need to Know (June 2026); Arnold & Porter eData Edge, Court Rules Expert's AI Prompts Are Fair Game Under Rule 26 (May 2026); Eugene Volokh, Reason (Volokh Conspiracy), May 26, 2026; E&E News by Politico, Climate lawsuit hints at how courts may treat AI.

Frequently Asked Questions

What did the court order in Conservation Law Foundation v. Shell?

On May 18, 2026, Magistrate Judge Thomas O. Farrish ordered the plaintiff to produce the generative-AI prompts its expert, Dr. Naomi Oreskes, used to narrow Shell's document production, holding they are part of her methodology and discoverable under Rule 26.

Is the order in effect now?

No. The district court stayed the order pending resolution of the plaintiff's Rule 72(a) objection, which District Judge Vernon Oliver will decide. As of early July 2026 the order is paused and the outcome is undecided.

Why were the prompts considered discoverable?

The court treated the AI-assisted culling of documents as an aspect of the expert's methodology, and an expert's methodology is generally fair ground for discovery under Rule 26.

Did the Rule 29 stipulation protect the prompts?

No. The court held that any agreement barring otherwise-relevant discovery must be quite clear, and that describing AI prompts as notes was not clear enough to shield them.

Does this mean all AI prompts are always discoverable?

No. This is a fact-specific trial-court order, now stayed and under review. It is an influential early decision, not binding precedent, and its reasoning could be narrowed or changed on the pending objection.