A recent decision from the U.S. District Court for the District of Connecticut addressed a discovery dispute concerning an expert witness's use of generative artificial intelligence (AI). The court, applying existing discovery rules and principles, required the disclosure of the expert witness’s AI prompts, the documents she uploaded to the tool, and the resulting AI outputs. The case, Conservation Law Foundation v. Shell Oil Co., (FN1) highlights the role of AI in expert preparation, methodologies, analysis, and opinions, and clarifies the parties’ obligations to disclose AI-related materials during discovery. Notably, the plaintiff’s expert incorporated an AI tool directly into her methodology before forming her opinions, thereby making her AI usage, prompts, inputs, and outputs subject to discovery.
Facts
The Conservation Law Foundation (CLF) filed a multi-count complaint against Shell Oil Company and others, alleging violations of the Clean Water Act and the Resource Conservation and Recovery Act at a Connecticut terminal. As part of their case, CLF’s expert and the expert’s assistant used an AI tool–specifically, ChatGPT on the Microsoft Azure platform–to analyze documents produced by the defendants. The expert and her assistant created prompts and uploaded some discovery documents to the AI platform, asking the tool to identify relevant materials. Crucially, neither the expert nor her assistant validated the AI outputs against the full set of documents, and there was confusion about which documents were uploaded and how many. CLF later admitted that only some of the discovery documents were uploaded, but refused to identify or specify which ones. During the expert’s deposition, the defendants asked about her methodology, including her prompts, the documents she shared with the AI tool, and the tool’s outputs. The expert was unable to answer these questions, deferring instead to her assistant, who, she indicated, might possess the necessary information. CLF subsequently asserted that the prompts and AI outputs had not been preserved. The defendants filed a motion to compel (ECF 941), arguing that this lack of disclosure violated the expert discovery rules and deprived them of their right to cross-examination.
The case was further complicated by a standing protective order that limited the use and disclosure of the defendants’ confidential and proprietary information produced during discovery. Despite the protective order, the plaintiff’s expert uploaded some of the defendants' proprietary information to the third-party AI platform. The defendants argued that uploading their proprietary information to an external AI system exposed it, raising concerns about potential breaches of confidentiality. The defendant further noted that the plaintiff failed to provide documentation or assurances confirming that the AI tool’s security settings precluded data retention, logging, or model training using their proprietary information. There was a standing agreement between the parties to exclude certain materials, such as expert notes, drafts, or communications, from disclosure, which plaintiffs argued excluded them from providing these assurances or producing the prompts and AI outputs.
Court’s Ruling
The court granted the defendants’ motion to compel, holding that AI prompts and queries used by an expert are part of the expert’s methodology and are therefore discoverable under Federal Rule of Civil Procedure 26(b). The court rejected the plaintiff’s argument that a prior agreement excluding “expert notes” shielded these materials, finding the agreement insufficiently clear to override the relevance of the AI-related materials. The court ordered the plaintiffs to revise their discovery responses and either produce any relevant AI prompts or queries or formally certify that none exist. The court cautioned that failure to comply could result in sanctions. The court reasoned that the AI prompts in this case were inherent to the expert’s methodology, making her AI prompts and outputs discoverable. The court rejected the plaintiff’s argument that the agreement between the parties excluded them from providing this discoverable material because the agreement was not specifically clear regarding the shielding of AI materials.
Discovery and Depositions in the Age of AI
As reported by the Maryland State Bar Association, one of the most practical impacts of AI tools on lawyers is how they conduct discovery and retain and cross-examine experts. See AI’s Immediate Impact on Case Preparation, Discovery, and Depositions, Langham, Pamela, MSBA Website (June 24, 2024).
In litigation, lawyers must inquire through interrogatories or depositions if an expert has used or reasonably relied upon AI in formulating their assessments and opinions. Specifically, lawyers should determine whether any portion of the information that was deemed reliable by the expert was generated by AI technologies. The necessity of further probing is also evident when an expert denies using AI at any stage of their methodology. For example, it is crucial to probe the basis of a medical expert’s denial of use of AI, particularly if a medical diagnosis–such as identifying a herniated disc–was influenced by AI interpretation of diagnostic imaging. Moreover, if AI has been employed as an instrument in research, it is essential to determine whether the expert has independently verified the AI-generated research outcomes against established peer-reviewed literature. This line of inquiry extends to any witness or professional rendering expert opinions. Given the integration of AI into expert analysis, lawyers must diligently assess the extent of AI’s influence on their conclusions.
Conclusion
The court applied traditional discovery rules and principles in Conservation Law Foundation v. Shell Oil Co., holding that AI prompts, queries, and outputs used by an expert are part of the expert’s methodology and are thus subject to disclosure under the Federal Rules of Civil Procedure 26(b). The court’s approach reinforces the requirement of transparency in expert analysis, regardless of whether traditional or AI technological methods are used.
AI has permeated our society. Lawyers should anticipate the use of AI-assisted work product by their experts and ensure that all relevant materials are preserved, protected, and properly disclosed. Lawyers analyzing an opponent’s expert’s opinion should likewise anticipate that AI may have been used, and frame their discovery requests and cross-examination accordingly. These practical considerations must be factored into the decision to retain an expert witness and into the broader discovery process and strategy when drafting, propounding, or responding to discovery requests.
_______________
FN1 Conservation Law Foundation v. Shell Oil Co., Case No. 3:21-cv-00933, ECF 970 (D. Conn. May 18, 2026).