Expert Paid $475 an Hour Wrote 90% of Court Report with ChatGPT

An expert hired by the American corporation 3M to defend the company in court over a fatal industrial explosion decided to use ChatGPT. The neural network was not merely asked to help with the text: it was tasked with preparing arguments that were supposed to prove 3M’s complete innocence.

Later, the victims’ lawyers gained access to the expert’s correspondence with the chatbot. They obtained around 350 pages of dialogue, and the specialist himself admitted in court that approximately 85–90% of his expert report had been created with the help of ChatGPT.

The story began with a tragedy in Houston. On the morning of January 24, 2020, a powerful explosion occurred at the Watson Grinding & Manufacturing facility. Before that, propylene—a highly flammable gas—had accumulated inside the building. According to the federal investigation, the leak continued, while the gas detection and emergency shutdown system failed to warn workers of the danger. When the lights were switched on inside the building, the gas ignited. Three people were killed, dozens were injured, and around 200 nearby homes and other structures were damaged or destroyed.

The gas detection system became a separate issue in the subsequent court proceedings. 3M had been engaged to inspect and maintain it. The plaintiffs argued that the company had failed to identify or fix problems with the system, which was supposed to warn of a dangerous gas buildup.

It is important to note that the federal investigation did not assign direct responsibility to 3M. Investigators pointed to a whole range of violations and problems at the Watson Grinding facility itself. In court, the victims separately sought to prove 3M’s responsibility for work on the gas detection system.

To defend 3M, the company hired engineer Josh Autenrieth from Knighthawk Engineering. According to court documents, the company paid the firm around $90,000, while the specialist’s rate was $475 an hour. The case materials indicate that he worked at least 174 hours.

And this is where artificial intelligence entered the story.

While preparing his expert report, Autenrieth uploaded a large amount of case material to ChatGPT and explained to the neural network that he was serving as an expert for 3M. He then asked the chatbot to prepare an expert report defending the company.

One of the prompts was particularly revealing: the specialist asked ChatGPT to «show how 3M was 0% responsible» for the explosion at Watson Grinding. In another prompt, he asked it to create an expert report that would confirm that 3M had complied with the required operating standards.

ChatGPT did in fact prepare a draft report of approximately 30 pages. An early version even concluded that 3M was «0% responsible» for the explosion.

The expert then asked the neural network to review its own text through the eyes of a lawyer for the opposing side. ChatGPT pointed out that the categorical wording about «0% responsibility» could become an easy target during cross-examination and create the impression that the specialist was acting more like the company’s lawyer than an independent expert. That specific wording did not make it into the final version of the report.

But the problems were only beginning.

During discovery, the plaintiffs’ lawyer Will Moye found a five-page document titled Citation Overlay, which he believed had been created with the help of AI. He demanded that the expert’s prompt history be disclosed.

During his questioning, the hearing even had to be suspended for approximately three hours, after which the parties were provided with around 350 pages of Autenrieth’s correspondence with ChatGPT. The materials documented in detail the specialist’s work on the future expert report.

The correspondence showed that the expert used the neural network not only to edit an already completed text. He uploaded documents, asked it to formulate arguments in 3M’s favor, analyze weak points in the report, assess its quality, and even tried to use ChatGPT to identify equipment in photographs.

In one of the dialogues, the specialist uploaded an image of a gas detector—equipment directly relevant to the dispute—and asked the neural network what exactly was shown in the photograph.

In court, Autenrieth did not deny actively using AI, but he disagreed that ChatGPT had effectively replaced him as the expert. He argued that his professional knowledge determined the substance of the report, while artificial intelligence helped create a draft and develop the arguments. The specialist also referred to more than 20 years of experience in the industry.

However, even by the expert’s own assessment, 85–90% of the final report was created by ChatGPT. The plaintiffs’ lawyer argued that this was no longer ordinary use of AI as an editor or support tool, but the neural network’s actual participation in forming the expert opinion.

Ultimately, the ChatGPT episode became one of the most unusual aspects of the years-long litigation.

On August 11, 2026, a Harris County jury in Texas awarded $61.5 million to 24 victims of the explosion. The jurors found 3M responsible for 30% of the damages, assigning the remaining 70% to Watson Grinding.

3M disagreed with the verdict, saying it considered the decision unfounded and intended to appeal it.

At the same time, the latest verdict is far from the end of the story. More than 1,900 other claims by people injured in the explosion remain as part of the extensive litigation against 3M. Jurors had previously issued other multimillion-dollar awards against the company, although 3M managed to have itself found entirely free of liability in one earlier trial.

The tragedy occurred more than six years ago, but its consequences continue to be resolved in court. The ChatGPT episode has added another important question to the proceedings: where does an expert’s use of artificial intelligence end and the actual transfer to a machine of work for which the specialist is paid and professionally responsible begin?

In this case, the specialist was paid hundreds of dollars an hour specifically for expert work. Yet, by his own admission, a chatbot created a significant portion of the report he prepared. The irony is that the neural network helped write 3M’s defense and then warned the expert itself that an overly categorical claim of zero responsibility could look like an advocate’s position.

However, changing the wording proved insufficient. At the end of the process, the jurors still decided that 3M bore 30% of the responsibility and awarded the victims $61.5 million.