An expert witness retained by 3M used ChatGPT to draft the core of his defense report in litigation stemming from the deadly January 24, 2020, Watson Grinding & Manufacturing explosion in Houston, according to court records uncovered by plaintiffs’ attorneys. Josh Autenrieth, an engineer with KnightHawk Engineering, submitted prompts instructing the artificial intelligence to show how 3M met standards of care and was zero percent at fault for the industrial disaster, which killed three people, injured dozens, and caused roughly $200 million in property damage.
Discovery Uncovers 350 Pages of ChatGPT Prompts
During trial preparations in Harris County, plaintiffs’ attorney Will Moye discovered a five-page document titled “Citation Overlay” that displayed formatting characteristic of AI generation. According to court disclosures, Moye demanded the complete conversation history behind the document, pausing a deposition for three hours. The resulting disclosure produced 350 pages of ChatGPT transcripts, public chat links, and drafting prompts.
The transcripts revealed that Autenrieth uploaded hundreds of pages of court records and asked the language model to generate a 30-page expert report. Beyond drafting, the engineer used the AI to simulate opposing counsel, identify potential weaknesses in the defense, and analyze photographs of gas detectors—the central hardware at issue in the litigation. Legal assessments by both sides at trial indicated that between 85% and 90% of the final filed expert report originated from ChatGPT.
The 3M Gas Detection Controversy
The underlying litigation focused on liability for the massive propylene gas leak that triggered the northwest Houston blast. Investigators with the U.S. Chemical Safety Board traced the initial leak to a poorly crimped rubber welding hose that leaked roughly 600 gallons of gas overnight before a worker’s arrival touched off an explosion via a light switch.
Civil lawsuits targeted 3M because the company held a contract to inspect and maintain the facility’s gas detection system. Plaintiffs alleged that an “open loop fault”—a severed connection between the gas detector and the industrial control computer—rendered the safety system inoperable, and that 3M failed to identify or report the failure.
AI Flags Its Own Advocacy Risk
According to the chat transcripts, ChatGPT initially drafted language explicitly stating that “3M is 0% responsible for the January 24, 2020 explosion.” When Autenrieth directed the AI to review the text from the perspective of opposing counsel, the model warned that the absolute phrasing served as an easy target that would expose him as an advocate rather than an objective expert.
The human expert accepted the AI’s critique and removed the explicit percentage from the final court filing. Despite the AI-assisted defense, juries in Harris County repeatedly assigned liability to 3M. Previous bellwether trials resulted in verdicts of $37.9 million and $118 million, with 3M assigned 49% fault in the latter case. A subsequent trial awarded $61.5 million with 3M assigned 30% of the fault, though a separate 2026 trial cleared the company entirely.
Worth a look