Connecticut Court Rules AI Prompts Discoverable as Expert Methodology

A U.S. District Court for the District of Connecticut has ruled that AI prompts used by an expert witness are discoverable, treating the specific instructions given to artificial intelligence as a core component of a professional's methodology.

In the case of Conservation Law Foundation, Inc. v. Shell Oil Company, the court compelled a testifying expert to produce the AI prompts and queries used to filter and identify documents. The ruling relied on Federal Rule of Civil Procedure 26, which governs the scope of discovery in civil litigation. In reaching this decision, the court explicitly rejected arguments that these prompts fell outside the scope of Rule 26 or were protected by a discovery-limiting agreement.

The Legal Framework

This decision centers on the transparency of expert testimony. Under Federal Rule of Civil Procedure 26, parties are entitled to information regarding the data and methodology an expert relies upon to form their opinions. Historically, this included spreadsheets, handwritten notes, or software parameters. By extending this requirement to AI prompts, the court has signaled that the "input"—the specific phrasing and constraints provided to a Large Language Model—is as critical to the final result as the output itself.

Why It Matters

This ruling challenges the notion of AI as a "black box" in legal proceedings. For years, the complexity of AI algorithms provided a layer of opacity, but this decision shifts the focus from the algorithm's internal weights to the human's specific instructions. If prompt engineering is deemed a core part of an expert's methodology, it forces a new level of scrutiny on how AI is used to synthesize evidence. Experts can no longer rely on the perceived neutrality of an AI tool if the prompts used to guide that tool are biased or overly narrow.

What's Next

Legal practitioners are now watching to see if this precedent spreads to other jurisdictions and whether it will lead to the creation of "prompt logs" as a standard requirement for expert disclosures. While this specific case focused on document filtering, future litigation may test whether prompts used for qualitative analysis or predictive modeling are similarly subject to discovery. For now, the ruling establishes that the bridge between a human expert and an AI's conclusion—the prompt—is a discoverable piece of evidence.