Trends in Expert Witnessing: AI Under the Microscope

By Noah Bolmer 

As generative AI finds its way into expert work, questions are arising regarding how experts use these tools and disclose their role in developing an opinion. Those questions take on a different dimension when opposing counsel seeks records associated with an AI tool during an engagement. A discovery dispute can expose parts of an expert’s working process that never appear in the final report, raising questions about which materials must be produced and which remain protected. Recent disputes offer a closer look at how courts and litigants are addressing these issues. 

Always check with your engaging attorney before using any AI tools. 

The Discovery Dispute 

In Villanueva v. Las Vegas Metropolitan Police Department, a dispute over an expert’s AI records arose during discovery in a civil rights case. Jose Villanueva sued the department and an officer over alleged violations of state and federal law during his detention at the Clark County Detention Center. Villanueva’s use-of-force expert, Tom Melton, testified at his October 1, 2025, deposition that he had used ChatGPT to assist with research and drafting his expert report. 

The defendants had requested the data and other information Melton considered in forming his opinions. After learning of his AI use during the deposition, they sought his ChatGPT history log. The plaintiff opposed production, arguing that the log contained protected expert work product and trial-preparation material. The magistrate judge ordered production of the full log from the bench. The plaintiff challenged that decision, bringing the dispute before U.S. District Judge Anne R. Traum. Her September 29, 2026 order addressed whether the ChatGPT history should be produced in full or whether some portions were protected from disclosure. 

What the Court Ordered Produced 

Judge Traum partially sustained Villanueva’s objection to the order compelling production of Melton’s ChatGPT history. After reviewing the log in camera, she distinguished between information Melton considered in forming his opinion and material reflecting his preparation and thought process. The analysis turned on Federal Rule of Civil Procedure 26, which requires disclosure of facts or data considered in forming an expert opinion while protecting certain draft reports and litigation-preparation materials. 

The court ordered production of all AI-generated responses summarizing case law that Melton had consulted. Under Rule 26(a)(2)(B)(ii), the relevant question was whether he had considered the material, not whether he ultimately relied on it. Because Melton consulted the summaries while developing his opinion, the court treated them as facts or data subject to disclosure. It also noted that the history log was the only place the defense could examine concerns about potentially hallucinated case citations, quotations, and broad legal standards. 

The court reached a different conclusion about Melton’s prompts. As Judge Traum explained, “a question is not something that an expert can ‘consider’ or ‘rely’ upon as ‘facts or data.’” She also found that the prompts could reveal his mental impressions, litigation strategy, and the issues he considered important. The court protected the portions of the log relating to preparation for deposition and the creation of draft report material. It found that the latter fell within Rule 26(b)(4)(B), which protects draft expert reports “regardless of the form in which the draft is recorded.” 

The court also found that Melton’s use of ChatGPT as a research, drafting, and editing tool did not waive any otherwise applicable privilege in this case. That conclusion was specific to the circumstances before it; it does not establish that all AI-assisted work retains privilege. 

The decision is a federal district court discovery ruling from Nevada, not a nationwide rule governing AI-assisted expert work. Its significance lies in how it applies existing discovery protections to particular kinds of AI-generated material. The court did not establish a general test for every AI history or decide the admissibility of AI-assisted expert opinions. It noted that “many of Defendant’s concerns about Melton’s methodology and reliance on ChatGPT are appropriate to address at trial as matters of impeachment.” 

When AI Use Draws Attention at Trial 

The Watson Grinding explosion litigation offers a different example of AI-assisted work coming under scrutiny. 3M retained an engineering expert to address the standard of care for servicing the facility’s gas detection system, which plaintiffs alleged had not been properly maintained. According to 404 Media’s reporting, the expert’s ChatGPT prompts included a request to “show how 3M is 0% at fault for the explosion at Watson Grinding.” Plaintiffs’ counsel discovered a document tracing the sources used in the draft report and sought the underlying AI records. 

The records became central to questioning about how the report had been prepared. According to CBS News, the expert initially testified at his deposition, “AI did not write my report, that’s incorrect,” before acknowledging its role in drafting and formatting. After 3M removed him from its witness list, the plaintiffs subpoenaed him to testify at trial about his AI use. There, he defended the process, saying, “I don’t believe that was wrong to do because I had an opinion already.” 

The jury found 3M negligent and awarded the plaintiffs $61.5 million. The available reporting does not establish that the AI use determined the verdict, and no reported judicial ruling addressed whether the expert’s use of ChatGPT was improper. The episode nevertheless illustrates how prompts and working records can become a focus of trial examination, giving the opposing side material to challenge the development and independence of an expert’s opinions. 

Practical Considerations 

The recent disputes offer several considerations for experts incorporating AI tools into their work: 

• Discuss AI use with retaining counsel before beginning work. Ask whether counsel has policies governing AI tools, which applications are approved, and whether there are restrictions on entering case materials or confidential information into them. 

• Treat AI interactions as potentially discoverable. Prompts, source materials, and generated responses may become relevant to discovery, depending on how they were used. Do not assume that an interaction is protected simply because it occurred in a private account or was part of an expert’s working process. 

• Verify AI-generated research against original sources. Check case citations, quotations, technical references, and factual assertions against authoritative materials. An AI-generated summary should not become the basis for an opinion without independent verification. 

• Be prepared to explain how AI contributed to the work. An expert should be able to distinguish between using a tool for research or drafting assistance and relying on its output in developing an opinion. The distinction may matter when the work is examined in discovery or at trial. 

Conclusion 

AI-assisted work complicates the boundary between an expert’s finished opinion and the process used to develop it. Villanueva shows how that distinction can affect discovery, while the Watson Grinding trial illustrates how an expert’s use of AI can become a subject of public examination. As these disputes continue to develop, the records behind an AI-assisted opinion may matter as much as the finished report. 

If you’d like to be considered for expert witness opportunities, join Round Table Group. For over 30 years, we’ve connected litigators with the most qualified experts across disciplines. Call us at 202‑908‑4500 or sign up today to learn more.