AINEWS
The 350 pages of Josh Autenrieth's ChatGPT prompts, including one asking it to show 3M was 0 percent at fault in the 2020 Houston Watson Grinding explosion that killed three, are in the court record.
When a Houston judge ordered an expert witness defending 3M to turn over his ChatGPT history, the 350 pages that came back did more than embarrass Josh Autenrieth. They put AI prompt logs into the legal discovery record, with the model's own self-critique functioning as a credibility test on the expert's theory of the case.
Autenrieth runs Knighthawk Engineering and bills himself as a twenty-year gas detection veteran. 3M's legal team hired him to defend the company in wrongful-death and liability suits arising from the January 24, 2020 Watson Grinding explosion in Houston. That blast killed three people and destroyed hundreds of homes. He was supposed to be the technical authority on whether 3M's products met the standard of care.
Instead, according to a report by 404 Media republished by Futurism, much of his expert opinion appears to have come from ChatGPT. The plaintiffs' attorneys suspected AI use during discovery. They asked the court to order production of his prompt history. 3M's lawyers complied. What they handed over was 350 pages of prompts and outputs, with public links to his full conversations.
His first prompt asked ChatGPT to "create an exceptional expert witness report defending the standard of care at 3M" and to show "how 3M is 0 percent at fault." The model produced a 30-page report that declared, in a passage quoted by 404 Media: "From a technical and standard-of-care standpoint, 3M is 0 percent responsible for the January 24, 2020 explosion."
When Autenrieth ran the draft back through ChatGPT and asked the model to critique its own work, the AI flagged the "0 percent responsible" line as "an easy target" for opposing counsel and recommended softening the language. The line was later removed from the version filed in court, but the model's critique of the removed line is now part of the discovery record.
In another exchange, Autenrieth uploaded a photograph of a gas detector and asked the model, according to the 404 Media reporting carried by Futurism: "what am I looking at?" The detector in the photo is described in the case as central to the question of what 3M's product did and did not do. The man hired to opine on it could not identify it without asking the chatbot.
Will Moye, the plaintiffs' lawyer, told 404 Media via Futurism that the expert "relied on AI not as an assistive device, but exclusively relied on ChatGPT to form his opinions."
A court order can compel a litigant to produce the inputs that shaped an expert's opinion, not just the polished report. The expert's working file is no longer the expert's alone. Anyone who has used a large language model to draft, outline, or sanity-check a section of an expert report has now seen a preview of what opposing counsel may try to obtain.
The AI's pushback is on the record alongside the report. When ChatGPT told Autenrieth that "0 percent responsible" was a giveaway, the model was doing what a good cross-examiner would do: marking the weakest claim. An expert who drops the most exposed sentence from a draft still has the model's critique on file. The deletion does not erase the assessment.
For 3M, the underlying legal exposure is unchanged. The Watson Grinding plaintiffs have always alleged that 3M's product was at fault in a blast that killed three and displaced hundreds. What is new is that the expert retained to push back on that theory has been compromised, on the record, by the tool he used to build his report.
Courts have been wrestling with AI-generated filings for two years. Most of the attention has gone to lawyers who filed briefs full of fake citations. That problem is a hygiene issue: a model hallucinated, the lawyer did not check, and the court sanctioned.
The Houston case is a different category. The expert did not file a hallucinated brief. He filed an expert opinion that reads as if it was independently formed, while the working record shows it was solicited from a chatbot instructed to reach a foregone conclusion. The discovery order is what reveals the gap between the work product and the work that produced it.
If other judges follow this lead, the practical effect is simple: an expert who leans on an AI to draft, edit, or stress-test an opinion should expect opposing counsel to ask for the prompt log. The next motion over AI-assisted expert work will read like the one that landed Autenrieth's 350 pages in court.
The case against 3M still has to be proved on the underlying facts of the Watson Grinding explosion. The expert retained to contest those facts has already been discredited, by his own prompts, in the file where the case is being tried.