When are an expert witness’s exchanges with AI discoverable? (Villanueva v. Las Vegas Metropolitan Police Department)

9 October 2026

The issue of the discoverability of an expert witness’s use of artificial intelligence during the preparation of an expert report is beginning to find its way to the courts.

In Conservation Law Foundation, Inc. v. Shell Oil Co. (Case No. 3:21-cv-00933, D. Conn.), a US judge found that generative artificial intelligence (AI) prompts used by an expert witness in preparing her expert report were discoverable.

While the plaintiff, who retained the expert, argued, among other things, that AI prompts were not discoverable under applicable court rules, the judge held that “[a]n expert witness’s methodology is fair ground for discovery…and under the fact[s?] of this case, the process by which [the expert] culled down the defendant’s document production into a subset to be worked with is an aspect of that methodology.”

Under Canadian law, it is arguable that a similar decision could easily be made by a court because, in general, discovery questions related to an expert’s methodology used to support an opinion are permissible and the methodology used or followed by an expert is important to the weight to be given to the expert’s opinion.

However, in Villanueva v. Las Vegas Metropolitan Police Department, (Case No. 2:24-cv-00125-ART-DJA), the United States District Court in Nevada determined that not all AI prompts used by an expert are discoverable or must be produced.

In Villanueva, the plaintiff claimed damages in connection with breaches of state and federal law that occurred while he was detained at a detention centre.

During the proceeding, the defendants brought several motions were, including a motion related to discovery.

On the discovery motion, the defendants sought the production of an expert witness’s ChatGPT history log because the expert testified at a deposition that he had relied on ChatGPT to assist him with research and drafting his report.

The defendants’ motion was granted on the grounds that the history log was producible under Federal Rules of Civil Procedure 26(a)(2)(B)(ii) and (iii).

The history log included legal case summaries, questions connected to the expert’s mental impressions of the case and questions designed to assist him with his deposition. The legal case summaries were related to locating and retrieving cases dealing with the use of excessive force.

In ordering the disclosure of the entire history log, the motion judge found that it was not work product and that solicitor-client privilege did not protect the log from disclosure.

The plaintiff objected to this finding and argued that rule 26(a)(2)(B)(ii) limited disclosure to “facts or data considered”.

On review of the motion judge’s decision, the District Court explained that under the applicable rules, a party could seek discovery regarding any nonprivileged matter that was relevant to a party’s claim or defence and proportional to the needs of the case.

With respect to the disclosure of the case summaries, the court found that the summaries constituted “facts or data” that the expert considered and were producible.

However, with respect to other aspects of the ChatGPT history log, the court concluded that the motion judge had clearly committed an error. The questions and prompts, the use of ChatGPT to create a deposition outline and the creation of a draft report were, contrary to the motion judge’s ruling, all protected.

The court reasoned that questions and prompts could not be “facts or data” because questions and prompts could not be “considered” or “relied” upon by an expert.

Similarly, the questions and prompts related to the creation of a deposition outline were not producible because a deposition outline was not “facts or data”. Furthermore, the court explained that an expert witness was entitled to invoke discovery privileges under the rules. The court stated that the deposition outline was protected by trial preparation privilege under rule 26(b)(3)(A) and was protected under rule 26(b)(3)(B) which provided that it was impermissible to disclose “mental impressions, conclusions, opinions, or legal theories.”

Rule 26(b)(3)(A) provided that a qualified privilege also protected the disclosure of “mental impressions, conclusions, opinions, or legal theories of a party’s attorney or other representative concerning the litigation.” (Emphasis added)

Lastly, the court found that rule 26(b)(4)(B) protected the expert’s draft reports. Although the expert had testified that he had copied and pasted ChatGPT outputs into his own report, under the rule, expert drafts were protected “regardless of the form in which the draft is recorded.”

Importantly, the court commented on an objection about whether the expert’s use of ChatGPT waived privilege. Although the motion judge made no decision on whether the history log was privileged or whether the expert’s use of ChatGPT had waived privilege, the court stated that using ChatGPT for the purposes of research, drafting and editing did not result in the waiver of the privilege.

The court explained “…ChatGPT is not a person, but was an application…”

The key takeaway is that this case may inform how Canadian courts approach the discoverability of an expert’s interactions with generative AI when preparing an expert report. While concerns have been raised in many circles that the use of the ChatGPT by professionals can result in the waiver of privilege, the findings in this case may suggest otherwise. However, this question will likely be the subject of continuing legal debate as courts continue to grapple with the growing use of AI by lawyers and experts.

[This blog was not written with the use of artificial intelligence.]

 

(This blog is provided for educational purposes only, and does not necessarily reflect the views of Gardiner Roberts LLP).