3M expert told ChatGPT to build zero-fault case before $61M explosion verdict

Why it matters

AI-assisted expert work can create a discoverable record of how conclusions were formed. In this case, the prompts exposed an outcome-first assignment inside litigation involving three deaths and a $61M verdict.

An expert witness retained by 3M used ChatGPT to draft most of his report for a civil trial over a fatal Houston industrial explosion, feeding the chatbot an explicit objective: build a case that left 3M with no responsibility.

Jason Koebler (@jason_koebler) and 404 Media reported on August 17th that Josh Autenrieth, an engineer with Knighthawk Engineering, asked ChatGPT to "show how 3M is 0% at fault" for the January 24th, 2020 explosion at Watson Grinding and Manufacturing. The prompts, draft report, court transcripts and discovery records expose how a generative AI session became part of the evidentiary record in litigation carrying tens of millions of dollars in potential liability.

Autenrieth joined Knighthawk as director of reliability and plant operations in November 2025, according to a trade publication announcement. His earlier engineering and reliability roles included work at Olin, Dow Chemical, Phillips 66 and Sherwin Alumina. At trial, he cited more than 20 years of industry experience and maintained that the conclusions were his own, with AI helping him construct a draft that he reviewed and changed.

The prompts became evidence

Plaintiffs' attorney Will Moye discovered what 404 Media described as a five-page "Citation Overlay" that appeared to have been generated by ChatGPT. Moye demanded the underlying conversations during Autenrieth's deposition. Proceedings paused for three hours, after which the plaintiffs received about 350 pages of ChatGPT records, including public links to the conversations.

Those records showed an outcome-first process. Autenrieth told ChatGPT that he had been retained to defend 3M and asked it to create an expert report supporting 3M's standard of care. He uploaded court materials, asked the model to identify safety violations by Watson Grinding and instructed it to rebut another witness's conclusions.

ChatGPT generated a report of roughly 30 pages. An early version declared that 3M bore no technical or standard-of-care responsibility for the explosion. That sentence was removed before the report was filed after Autenrieth asked ChatGPT to assess the draft from opposing counsel's perspective. The model warned that an absolute finding would make the witness look like an advocate, according to the disclosed conversation.

Autenrieth also uploaded an image of a gas detector and asked ChatGPT to identify what he was viewing and name industrial gas-detector models. Moye told 404 Media that the equipment was central to the dispute. Autenrieth repeatedly submitted revised drafts for editing and asked the model to grade the report and anticipate attacks on his qualifications and conclusions.

The filed report retained the structure of the initial ChatGPT output, with substantial passages matching the model's drafts and revisions, 404 Media found. During examination at trial, Moye and Autenrieth characterized 85% to 90% of the submitted report as ChatGPT-generated. Autenrieth testified that he supplied the opinions, reviewed the output and changed material that did not reflect his views.

Court records cited by 404 Media indicate that 3M paid Knighthawk roughly $90,000 for its analysis. A 3M filing listed Autenrieth's rate at $475 an hour.

A $61 million verdict

The dispute stems from an explosion shortly before 4:30 a.m. on January 24th, 2020. The U.S. Chemical Safety and Hazard Investigation Board found that a hose disconnected inside a coating booth, releasing propylene that accumulated inside the Watson Grinding building and ignited. Two workers and a nearby resident died, and more than 450 surrounding structures were affected.

The federal investigation identified failures in Watson Grinding's process-safety management, emergency preparation and maintenance of its gas-detection and shutdown systems. Plaintiffs have argued that work performed on gas-detection equipment by Detcon, a former 3M subsidiary, also contributed to the blast.

The latest jury awarded the plaintiffs more than $61 million and assigned 30% of the responsibility to 3M and 70% to Watson Grinding. The verdict rejected the zero-liability result Autenrieth had directed ChatGPT to support.

The litigation remains much larger than one trial. In its quarterly filing for the period ending June 30th, 3M said more than 1,900 claimants had brought allegations against multiple defendants. Earlier juries reached conflicting conclusions: 3M was found partially liable in June and November 2025, then cleared of liability in an April 2026 trial. 3M appealed the first two judgments and said it had not recorded a liability because the losses were not probable or reasonably estimable.

AI work leaves a trail

Texas procedure gives opposing parties broad access to the foundations of a testifying expert's work. The state's current civil rules require retained experts to disclose documents, reports, models and data compilations provided to them, reviewed by them or prepared for their testimony. The Texas Rules of Evidence allow experts to be examined about their underlying facts and data, and provide that an opinion is inadmissible when those materials do not supply a sufficient basis.

The rules focus on an expert's qualifications, methods and factual foundation. Generative AI adds another record capable of exposing how the conclusion was developed. Here, the final report revealed only part of the process. The prompt history showed that the requested destination had been selected before ChatGPT began analyzing the evidence.