A recent ruling from a federal magistrate judge in Connecticut signals that courts may increasingly treat certain uses of generative AI as part of an expert's methodology, and therefore discoverable, rather than merely a research aid that can remain hidden. 

In climate litigation brought against Shell, Magistrate Judge Thomas Farrish ordered production of AI prompts used by a testifying expert, Dr. Naomi Oreskes, who relied on ChatGPT to help identify and to sort through a large collection of documents in ​the case. The court reasoned that because the AI-assisted process informed how the expert selected and reviewed relevant materials underlying her opinions, the prompts were relevant to understanding and evaluating the reliability of the expert's methodology. The court noted that in expert discovery, an opposing party is entitled to examine not only an expert's conclusions, but also the processes used to reach them, including AI tools that may have influenced the expert's review of the evidence. The environmental group that retained Dr. Oreskes is fighting the order, asking the District Judge Monday to halt the disclosure. 

The implications on expert discovery

While this particular decision is being challenged, it serves as a reminder that when AI plays a substantive role in an expert's work, the underlying prompts may be viewed as discoverable components of the expert's methodology rather than protected background materials. As courts are still developing the legal framework governing generative AI, there remains risk that decisions today regarding the use of AI could have unknown implications in future litigation.

Parties may increasingly seek discovery into AI prompts, outputs, and related workflows where generative AI is used to organize facts or evidence, conduct analysis, or otherwise contribute to the development of expert opinions.

And the broader implications

And although this decision arose in the context of expert discovery, the basis could well extend beyond to all discovery and other uses of AI. Where employees use AI tools to gather information, analyze documents, draft communications, or make business decisions, opposing parties may increasingly argue that prompts, outputs, and chat histories are relevant sources of discoverable information. 

Other cases have held that even using AI at “the direction of counsel” is insufficient to protect the records as work product.

The ruling also highlights practical questions regarding document preservation and governance. Organizations will need to decide whether AI should be sanctioned for business-use at all; and if so, how AI records should be subject to existing document-retention and litigation-hold procedures. 

Key takeaways

Future discovery protocols are likely to include detailed discussion of preservation and production of AI-prompts and chat logs.

Counsel should consider addressing AI use proactively through employee policies, expert engagement letters, litigation-hold procedures, and attorney guidance. As AI becomes embedded in ordinary business and litigation workflows, organizations should assume that courts and adversaries will increasingly scrutinize not only AI-generated outputs, but also the prompts and interactions that produced them.