How Expert Witness AI Use Ends Up in Discovery

“Show how 3M is 0% at fault.” The absurd tale of how a 350-page ChatGPT history for 3M’s expert witness was produced during a deposition is more likely than you think. It’s an inevitable trend, and an unsettled question for federal courts as experts make generative AI part of their workflow.

Key Takeaways

• Expert witness AI use can be discoverable. In Conservation Law Foundation v. Shell Oil, a federal magistrate judge ordered production of an expert’s AI prompts and queries as part of the expert’s methodology under Rule 26. The order is stayed pending the district judge’s review.
• The fallout can be severe. 3M’s expert witness produced roughly 350 pages of ChatGPT history mid-deposition, and a Harris County jury returned a $61.5 million verdict, allocating 30 percent of the fault to 3M.
• Detection is getting easier. Under Article 50 of the EU AI Act, Google and Anthropic are watermarking AI-generated text, which could indicate the likelihood that a specific model produced a given document.

Expert Witness AI Use Exposed at Deposition

In recent litigation over the deadly 2020 explosion at the Watson Grinding and Manufacturing facility in Houston, Josh Autenrieth was an engineer 3M retained as an expert witness. At his deposition, the expert witness had his AI use exposed and eventually admitted to using ChatGPT extensively while preparing his standard-of-care report. Plaintiffs’ counsel in that case, Will Moye, says the expert’s AI use was revealed by a five-page “Citation Overlay” buried in a document dump produced the day before the expert’s deposition. In a sign of the times, Moye referenced his use of Claude as he described how a “quick scan” of the 14,000-page production uncovered the buried GPT output.

The ensuing mid-deposition production included approximately 350 pages of ChatGPT history. That history is what litigators dream of, including requests to essentially create “an expert witness report to defend 3M’s standard of care” and “show how 3M is 0% at fault.” Naturally, the deposition was not the end of that story. At trial, Plaintiffs’ counsel called that expert as a witness and took the opportunity to describe how the expert had purportedly “uploaded 200 plus case documents into chat,” and signed an expert report that was “90 percent… written by AI.” A Harris County jury ultimately returned a $61.5 million verdict and allocated 30 percent of the fault to 3M.

While that episode arose in Texas state court and the court did not compel disclosure of the expert’s AI chat history, it raises a broader question federal courts are now confronting: when a testifying expert uses generative AI, which aspects of that data use are discoverable?

This Is the Tip of the Iceberg

Just one week before 3M’s expert witness was grilled about his AI use, Plaintiff’s counsel in a District of Connecticut case, Conservation Law Foundation, Inc. v. Shell Oil Co., filed an emergency motion to stay an order to disclose their expert’s AI prompts. The magistrate judge considered the expert’s use of AI, described by Plaintiff’s counsel as “a specific license program that uses ChatGPT,” and ordered production of the prompts and queries used by the expert, Dr. Naomi Oreskes:

CLF first protests that artificial intelligence prompts used by an expert witness are not within the scope of discovery under Rule 26(b), but the Court disagrees. An expert witness’s methodology is fair ground for discovery; see, e.g., Macchia v. ADP, Inc., 711 F. Supp. 3d 162, 167-68 (E.D.N.Y. 2024), and under the facts of this case, the process by which Dr. Oreskes culled down the defendants’ document production into a subset to be worked with is an aspect of that methodology.

(Id., ECF Doc No. 970). That order was stayed while the district judge considered the plaintiff’s Rule 72 objection, so that won’t be the final word on that dispute. Even so, it foreshadows what will undoubtedly be a recurring theme in future discovery battles: the nature of an expert’s AI use will shape the Rule 26 analysis that follows, and a prompt that determines which evidence the expert “considers” in their report may itself be discoverable as part of the expert’s methodology. For the rare lawyer that knows where to look for an AI agent’s local session history, retrieving such evidence can be a roadmap to dismantling an adverse expert report.

This issue will become even more common as AI use becomes easier to detect because of another recent development: AI watermarking.

AI Watermarking and Other “Tells” Are a New Concern

On August 2, 2026, the European Union’s AI Act took effect, including Article 50 (“Transparency obligations for providers and deployers of certain AI systems”), which requires providers of generative systems to make synthetic text and other outputs machine-readable and detectable as artificially generated or manipulated, subject to technical-feasibility and editing exceptions. Google says its SynthID technology watermarks text generated through the Gemini app and web experience, as well as images, audio, and video. Anthropic announced in August that future Claude models will watermark text using a version of Google’s approach.

Such watermarks are not a smoking gun for AI-generated content, but they will offer a percentage likelihood (e.g., “90% likely”) that a specific large language model generated the provided text. Much like the “Citation Overlay” in the 3M case, it may serve as compelling evidence of AI use that could culminate in devastating admissions about the basis of an expert’s report.

As Apple discovered recently in Taction Technology, Inc. v. Apple Inc., you need that kind of strong evidence before pressing an expert regarding suspected AI use. Instead, Apple argued that the expert’s “increased use of the em dash in her corrected report” was evidence that she had used AI to generate it. The Court did not like that, criticizing the “em dash” AI evidence as “unimpressive” and “unpersuasive.” Beyond vindication for lovers of the em dash, the Taction Technology decision is also important for its analysis of the expert’s “substantial participation” in drafting the report.

For a deeper dive, read Mr. Seelinger’s next article on discovering an expert witness’s documents: [Link to Second Article focusing on Rule 26]. The article analyzes recent cases on Rule 26, including what constitutes an expert witness’s “substantial participation” in generating draft reports that are typically exempt from disclosure. That topic will undoubtedly take on heightened significance now that experts are leaning on AI for their reports.

See also Mr. Seelinger’s earlier Firm Insight, Are Your Conversations with AI Shielded from Discovery? Courts Are Split.

Frequently Asked Questions

Is expert witness AI use discoverable?

It can be. In Conservation Law Foundation v. Shell Oil, the magistrate judge held that an expert witness’s methodology is fair ground for discovery under Rule 26(b) and ordered production of the expert’s AI prompts and queries. That order was stayed while the district judge considers the plaintiff’s Rule 72 objection, so it is not final.

What happened with 3M’s expert witness?

Plaintiffs’ counsel found a five-page “Citation Overlay” in a 14,000-page production the day before the expert’s deposition. The expert then produced approximately 350 pages of ChatGPT history mid-deposition, including prompts asking for a report showing “how 3M is 0% at fault.” At trial, plaintiffs’ counsel described the report as “90 percent… written by AI,” and the jury returned a $61.5 million verdict, allocating 30 percent of the fault to 3M. That disclosure arose in Texas state court, and the court did not compel it.

Can AI watermarks prove an expert used AI?

Not on their own. Watermarks such as Google’s SynthID offer a percent likelihood that a specific large language model generated the text, which may serve as compelling evidence of AI use. Weaker “tells” fare poorly: in Taction Technology v. Apple, the court called Apple’s argument that an expert’s increased use of the em dash showed AI use “unimpressive” and “unpersuasive.”

Talk to Us Before Your Expert’s AI Use Becomes the Story

The 3M deposition and the Shell order show how quickly expert witness AI use can become the central issue in a case, and watermarking will only make it easier to detect. If your business is heading into litigation that will turn on expert testimony, Chris Seelinger and Scarinci Hollenbeck’s Litigation attorneys can help. Schedule a consultation to discuss how expert witness AI use could affect your case.