What Surrounds “show how 3M
is 0% at fault” in a Saved Chat?

CBS News and 404 Media report that an expert witness retained on 3M’s side of the case entered that clause into an AI tool, ChatGPT in 404 Media’s account. 3M told CBS News he reached his own conclusions and “used AI as a tool to assist in preparing his report.”

By William Douglas Baker

Founding Editor

Inside a Texas lawsuit over a 2020 explosion that killed three people at a Houston manufacturing plant, CBS News and 404 Media report that an expert witness entered the clause “show how 3M is 0% at fault for the explosion at Watson Grinding” into an AI tool, ChatGPT in 404 Media’s account. CBS News identifies him as Autenrieth, director of reliability and operations at KnightHawk Engineering. A saved ChatGPT conversation holds a user message containing that clause.

The message opens “Reread all attachments and prompts and create an exceptional expert witness report defending the standard of care of 3M” and ends by asking the model to “show how 3M is 0% at fault for the explosion at watson grinding and how my background and experience is well suited to render this professional opinion”. Between those requests, it asks the model to illustrate what the message calls a lack of process-safety management and safety at the plant, and to defeat the “comments and unbiased assumption” of a witness from the opposing side, with an example about work someone is not authorized, trained or permitted to do. The production that prints this conversation is Bates-stamped “AUTENRIETH”; no court record read for this edition ties its messages to him.

In a further message before the draft came back, the user set out a view of the work’s scope: “the scope of work of 3M techs was limited to gas detector was never the PLC or the associated wiring or other equipment”. One more message follows; then a single message reads “FINISHED” and “Create the expert witness report draft now,” and then comes the draft. The user had first uploaded the case files in batches, telling the model to wait until “FINISHED” was typed.

The draft is headed “DRAFT EXPERT WITNESS REPORT”, and the line beneath the heading names Autenrieth. The draft’s Assignment section says, in the first person, “I have been asked to evaluate whether 3M, through its gas detection service technician, met the applicable standard of care”. Within that section it also says “I understand that legal fault allocation is for the trier of fact.” and gives its technical and standard-of-care opinion that 3M “bears 0% technical causative responsibility” for the explosion. On our reading, the instruction states a result to be shown, and the draft’s Assignment section opens by describing its task as an evaluation before stating its opinion in the same section; this edition takes no position on whose opinion either was. No report as filed or served was among the files this edition read through October 14, 2026.

The ninth and last of the draft’s Final Opinions states, “From a technical and standard-of-care standpoint, 3M is 0% responsible for the January 24, 2020 explosion.” Its last line reads “Draft prepared for attorney and expert review.” Above the Law reports, in its own voice and without naming a source, that a “0% responsible” line “got cut from the final report”; it does not say which draft or which sentence it means.

Two statements to CBS News bear on this sequence. 3M’s statement says he “used AI as a tool to assist in preparing his report.” His lawyer’s statement says, “He did not delegate the formation of his opinions to ChatGPT, nor did AI replace his professional judgement.” Both are quoted below as CBS News reports them.

What the expert, his lawyer and 3M say

His lawyer told CBS News in a statement that “He did not delegate the formation of his opinions to ChatGPT, nor did AI replace his professional judgement,” and that “recent public characterizations offer a selective distortion of the record.” CBS News reports the lawyer told it that while Autenrieth did use AI to assist with formatting, structuring and drafting, he did so after reviewing the thousands of pages of materials given to him. Above the Law reports that Autenrieth “even said” that “AI helped me to draft a straw man to build off of,” and does not say when or where.

Autenrieth was an expert witness retained on 3M’s side of the case. CBS News reports he was “hired by 3M just weeks before the trial began”. CBS News reports that 3M said in a statement that he was an independent third-party witness, not a 3M employee, and was retained by outside counsel to provide his own analysis and opinions. The same statement says, “After independently reviewing the facts of the case and reaching his own conclusions, he used AI as a tool to assist in preparing his report.” CBS News also states that Autenrieth “did not respond to our email and calls for comment.”

At his deposition he testified, “AI did not write my report, that’s incorrect,” and moments later said it had helped with “formatting and everything else,” according to transcripts cited by CBS News. At trial, CBS News reported, he “defended his usage of AI,” testifying, “I don’t believe that was wrong to do because I had an opinion already.” The record keeps the instruction. Whose judgment a signature certifies is the authority question.

How the prompts surfaced

404 Media states that court transcripts, deposition documents and discovery records were shared with it. The prompts, which 404 Media says were provided in the case, are a 365-page set of 12 shared ChatGPT conversations, each page Bates-stamped “AUTENRIETH,” posted on DocumentCloud and linked from 404 Media’s article. CBS News reports that after the deposition 3M’s attorneys removed the expert from their witness list, and that the plaintiffs’ attorneys subpoenaed him to testify about the various prompts he entered into ChatGPT. Declassified has printed AI prompts before: No. 094 printed one that opens with a question, and No. 102 printed queries that ask for a description. A scan of Nos. 001 through 103 and the No. 104 draft, as of October 14, 2026, found no earlier edition that prints a drafting instruction stating the result it asks the machine to show, set beside the draft the machine returned. No. 081 said “whether drafts or file histories existed was not established”; here a saved conversation exists.

What the sources report about the verdict

CBS News reported on September 18, 2026 that the jury found 3M negligent in its safety checks of the alarm systems and that the court awarded $61.5 million; 3M told CBS News it disagreed and intended to appeal. CBS News reports that one of the plaintiffs’ attorneys told it he believes the choice to bring the expert in to testify about the prompts had an impact on the outcome of the case. No source this edition read through October 14, 2026, with 404 Media’s article read only through its free introduction, reports a court ruling, sanction or exclusion concerning the expert’s AI use, or establishes that it changed the verdict.

What a signature has to be read against

On our reading, an expert’s authority is certified at the signature, and an AI-assisted draft can leave a second document beside the first: the instruction that preceded it. No signed or filed report was read for this edition. 3M’s statement to CBS News says he “used AI as a tool to assist in preparing his report”; his lawyer’s says, “He did not delegate the formation of his opinions to ChatGPT, nor did AI replace his professional judgement.” The entries below list the texts this edition compares.

The entries

Where the argument stops

This edition does not decide whose independently formed opinion the draft states. It cannot verify what the expert reviewed beyond the accounts of 3M and his lawyer and the draft’s own Materials Considered section, and it read no report as filed or served. The testimony reaches this edition only through CBS News and Above the Law; no deposition or trial transcript was read, and 404 Media’s article was read only through its free introduction. CBS News’s statement that he did not respond concerns its own request, and it does not say when that request was made.

Of the twelve conversations, one was read end to end, one in part, and the other ten were searched by keyword. No docket, order or verdict form was read through October 14, 2026. Disclosure: Declassified is researched and drafted with AI tools, including Anthropic’s Claude models, and an OpenAI model gave this edition’s evidence ledger and draft an outside read. Agentics builds AI agents that speak in its clients’ voices.

The mapping from this case to any other expert’s work is editorial.