A forensic engineering expert hired to defend 3M in a wrongful death lawsuit told ChatGPT to prove the company bore "0 percent" fault for a 2020 Houston explosion that killed three people. He then billed $475 an hour for the work.
Josh Autenrieth of Knighthawk Engineering submitted an expert report that he and the plaintiffs' attorney later agreed was "85 to 90 percent ChatGPT," the New York Post reported. 3M paid Knighthawk roughly $90,000 for the analysis. The revelation came after opposing counsel subpoenaed the AI chat prompts themselves, exposing the full scope of what the expert had asked the chatbot to produce.
The case puts a sharp point on a question courts have only begun to wrestle with: what happens when expert testimony, the backbone of complex civil litigation, is quietly outsourced to a large language model trained on internet text rather than grounded in independent professional analysis?
The underlying case stems from a 2020 explosion in Houston that killed three people. Families of the deceased brought a wrongful death suit against 3M. As part of its defense, 3M retained Knighthawk Engineering and its expert, Autenrieth, to prepare a forensic report on the cause and responsibility for the blast.
Expert reports in cases like these carry enormous weight. Juries rely on credentialed professionals to interpret physical evidence, engineering failures, and chains of causation that laypersons cannot evaluate on their own. A report that clears a defendant of fault can reshape settlement calculations or swing a verdict.
Plaintiffs' attorney Will Moye did not take the expert report at face value. His legal team subpoenaed the AI chat prompts behind the document, a discovery tactic that is still relatively uncommon but growing in significance as AI tools spread through professional work.
What Moye received was striking. Breitbart reported that opposing counsel obtained 350 pages of ChatGPT conversations, including public links to Autenrieth's full chat history, after a three-hour pause during the expert's deposition. The prompts revealed that Autenrieth had explicitly instructed ChatGPT to defend 3M and to show the company was "0 percent at fault."
That instruction is the crux of the problem. An expert witness is supposed to offer independent professional judgment. Telling an AI chatbot to reach a predetermined conclusion and then presenting the output as expert analysis inverts the entire purpose of expert testimony.
Moye drew a clear line between legitimate use of AI as an assistive device and what happened here.
"This expert relied on AI not as an assistive device, but exclusively relied on ChatGPT to form his opinions and write his report."
That distinction matters legally. Courts have increasingly scrutinized AI-assisted legal filings since a series of high-profile incidents in which attorneys submitted AI-generated briefs containing fabricated case citations. But the expert witness context raises different stakes. A lawyer who submits a sloppy brief embarrasses himself. An expert who submits a conclusion-driven AI report in a wrongful death case involving three fatalities risks corrupting the factual record on which life-and-death accountability depends.
Autenrieth billed at $475 per hour. 3M paid Knighthawk Engineering approximately $90,000 for the analysis. The question of what professional work product that fee actually purchased now hangs over the case.
One of the most consequential details is the mechanism that exposed the conduct. Moye's team successfully subpoenaed the ChatGPT prompt history, a move that required both legal creativity and a willingness to treat AI-generated content as discoverable work product rather than protected internal process.
The three-hour deposition pause that preceded the production of those 350 pages suggests the disclosure was not voluntary. Autenrieth and his team apparently needed time to gather and produce the chat logs after the issue surfaced during questioning.
For litigators on both sides of the bar, the precedent is significant. If AI chat histories are discoverable, every expert and every attorney who uses generative AI to draft substantive work product must assume that opposing counsel can obtain the full prompt-and-response chain. That changes the calculus for anyone tempted to use AI as a shortcut rather than a supplement.
Moye himself framed the incident as a cautionary tale that extends well beyond this single case.
"Every lawyer needs to make sure their own experts aren't generating work product in a way that's insincere."
That warning cuts in two directions. Defense counsel who retain experts have an obligation to ensure the opinions they sponsor in court reflect genuine professional analysis. And plaintiffs' counsel now have a proven roadmap for challenging expert reports they suspect were generated by AI: subpoena the prompts.
Courts have not yet established uniform rules governing AI use in expert reports. Some jurisdictions require attorneys to disclose AI-assisted filings, but those rules generally target briefs and motions, not the expert reports that underpin them. The gap leaves room for exactly the kind of conduct alleged here.
Authorities and the court have not publicly confirmed whether Autenrieth disclosed his use of ChatGPT to 3M before submitting the report. Investigators will need to determine whether 3M's legal team knew the report was AI-generated when they presented it, or whether they were also misled by their own retained expert.
The court has not publicly addressed whether the report will be excluded or whether sanctions will follow. Those procedural steps will shape how aggressively other courts police AI-generated expert testimony going forward.
Three people died in that Houston explosion. Their families filed suit seeking accountability. The expert report meant to shield 3M from that accountability now stands exposed as a document that was, by the expert's own agreement, almost entirely machine-generated to reach a predetermined conclusion.
The $90,000 fee makes the optics worse, but the real damage is to the integrity of the adversarial process. Expert testimony works only when experts do the work. When a credentialed professional tells a chatbot to write the answer and then signs his name to it, the courtroom loses the one thing it cannot function without: reliable, independent judgment.
If courts do not draw a hard line here, every expert report in every case becomes suspect, and the families who depend on honest litigation to hold powerful defendants accountable will pay the price.
