The filing

Conservation Law Foundation, Inc. v. Shell Oil Company et al., Case No. 3:21-cv-00933-VDO, U.S. District Court, District of Connecticut. Magistrate Judge Thomas O. Farrish’s order: May 18, 2026 (ECF No. 970). CLF’s emergency motion to stay the order: June 1, 2026 (ECF No. 974).

What happened

Shell moved to compel CLF to produce AI prompts used by its expert, Dr. Naomi Oreskes, who had used AI to process Shell’s discovery documents. CLF argued the prompts fell outside expert discovery under Fed. R. Civ. P. 26(a)(2)(B)(ii) because they were used only to filter documents, not form opinions. CLF also argued that a Rule 29 stipulation shielded the materials: the parties had agreed not to seek discovery of each other’s “expert notes, drafts, or communications needed by, and made during, the report drafting process,” and CLF contended the prompts qualified as “notes.”

Magistrate Judge Farrish disagreed on both counts, finding the Rule 29 language was not “quite clear” enough to bar otherwise-relevant discovery. He ordered CLF to revise its responses to include any AI prompts and queries used by Dr. Oreskes or her team, with Rule 37(b) sanctions available if the representation later proved untrue. CLF filed an emergency motion asking USDJ Vernon D. Oliver to stay the order pending review, on the grounds that no federal court has previously held an expert’s AI prompts are discoverable under Rule 26. USDJ Oliver granted the stay pending his consideration of CLF’s objections.

Why it matters

The ruling treats prompt engineering as part of an expert’s analytical method, not background process. If it holds, parties using AI to analyze documents in litigation will need to treat those prompts as potential discovery material from the outset.

Parties need to specify who created the prompts, which versions and settings to use, whether prompts must be produced or kept confidential, and how prompts and outputs will be stored, disclosed, or protected by privilege. Using precise language minimizes ambiguity and reduces the risk of sanctions.

In practice, this means thoroughly documenting AI workflows, including prompt provenance, version histories, model limitations, data sources, and decision rationales. Firms ought to obtain sworn confirmations of completeness and establish procedures for custodians to find prompts across systems. If your Rule 29 agreement covers AI-assisted discovery, be precise about scope, preservation, and production to avoid disputes, trial delays, and uncertain results.

JetStream’s take

CLF’s defense is that the AI only filtered documents, it did not form opinions. Proving that distinction requires the same prompts CLF says it never kept.

Source

ECF No. 974, CLF Emergency Motion to Stay, June 1, 2026 (Scribd embed)
ECF No. 970, Magistrate Judge Farrish’s Order, May 18, 2026, available on PACER, Case No. 3:21-cv-00933-VDO (D. Conn.)

Conservation Law Foundation's Emergency Order to Stay filed June 1, 2026 by JetStream AI Advisory