AI Use by Expert Witness Backfires in Texas Explosion Lawsuit

Artificial intelligence’s role in the legal system is facing renewed scrutiny after a multimillion-dollar lawsuit in Texas revealed that an expert witness used ChatGPT to generate more than 80% of his testimony report. The incident has raised alarms among legal professionals about the risks of relying on AI tools that may lack neutrality and objectivity.

The case centers on a catastrophic explosion at a Houston warehouse in January 2020, which claimed three lives and caused extensive damage to hundreds of homes and businesses. Plaintiffs sued the corporate manufacturing giant 3M, arguing the company was negligent in its supervision of safety checks at the facility.

Jason Autenrieth, the director of reliability and operations at KnightHawk Engineering, was hired by 3M’s outside counsel just weeks before the trial to serve as an expert witness. During depositions, plaintiffs’ attorneys discovered that Autenrieth had submitted 300 pages of prompts to ChatGPT while preparing his report. Among these prompts was a direct request asking the AI to “show how 3M is 0% at fault for the explosion at Watson Grinding.”

Adam Lewis, an attorney representing the plaintiffs, described his shock at the scale of AI involvement. While Autenrieth initially denied that the AI wrote the report, he later conceded it assisted with formatting and other elements. However, Lewis noted that the prompts included efforts to undermine opposing witnesses, suggesting the tool was used strategically rather than merely for drafting assistance.

Following the deposition revelations, 3M removed Autenrieth from their witness list. Nevertheless, plaintiffs’ attorneys subpoenaed him to testify about his specific interactions with the AI. During the trial, Autenrieth maintained that he already held his opinions and that using the tool was not improper.

The jury ultimately found 3M negligent in its safety inspections of alarm systems that could have prevented the blast, awarding the victims $61.5 million last month. A spokesperson for 3M stated they disagree with the verdict and intend to appeal, maintaining that the expert was an independent third party who used AI only as an assistive tool after reviewing thousands of pages of case materials.

Legal scholars warn that such cases highlight broader dangers. Damien Charlotin, an adjunct professor at the University of California College of the Law in San Francisco, noted there have been over 1,300 documented instances of AI hallucinations in U.S. legal proceedings since 2023. These errors often involve fabricated citations or fictitious studies that appear plausible.

Maura Grossman, a research professor at the University of Waterloo, emphasized that AI models are inherently “sycophantic” and designed to please the user rather than provide neutral analysis. She urged firms to exercise caution, ensure full disclosure of AI usage, and prioritize fact-checking. “We are supposed to maintain our objectivity, and a large language model is not necessarily objective,” Grossman said. “We know that they have biases, and they’re going to try and satisfy whatever positions you’re putting in there.”