An AI Told to Clear 3M of Blame Wrote the Expert Report That Went to Court
What happened
A 2020 explosion at a Houston industrial facility killed three people and destroyed roughly 200 homes. The victims' families sued, and 3M, named as a defendant, retained an expert to analyze who bore responsibility. According to reports from the proceedings, that expert, Josh Autenrieth, submitted his analysis after prompting ChatGPT to show that 3M was "0% at fault." He then filed the result as expert evidence.
Expert testimony in civil litigation carries a specific weight: it is supposed to represent the independent, professional judgment of a qualified analyst. Courts treat it differently from lay witness accounts precisely because experts are assumed to derive conclusions from evidence, not work backward from a requested outcome. What Autenrieth reportedly submitted reversed that premise entirely. The analysis was framed as expert opinion but was substantially drafted by a language model instructed from the outset to reach a particular verdict.
The specific prompt matters here. Asking an AI system to demonstrate that a party is "0% at fault" is not analysis. It is a brief with a conclusion already attached. A language model given that instruction does not weigh the evidence, identify the counterfactuals, or flag uncertainty. It constructs arguments in the direction it was pointed. When the result was submitted to court as expert analysis, plaintiffs were effectively handed a document with no independent analytical basis, though it carried the formal standing of one.
The submission prompted extensive discovery and trial scrutiny. Once the process behind the report became visible, opposing counsel had grounds to challenge not just the conclusions but the methodology underneath them. An expert report that cannot survive a basic question about how it was produced undermines the party that filed it, and in this case that cost fell on 3M's litigation position at exactly the moment when causation needed to be established clearly.
The deeper problem this incident names is not that an expert used an AI tool. It is that courts, at the moment a report is filed, have no reliable mechanism to verify what actually produced it. An attorney's certification covers the expert's signature, not the provenance of every paragraph. A provable record of what a system did, what inputs it was given, and what instructions shaped its output would have surfaced the prompted conclusion before it reached opposing counsel as a surprise. Without that record, any party to technical litigation faces the possibility that the expert on the other side is a prompt with a credential attached.
Reported impact
- Affected parties
- Not publicly disclosed
- Harm type
- Not publicly disclosed
- Scale
- Not publicly disclosed
- Financial impact
- Not publicly disclosed
- Regulatory action
- Not publicly disclosed
Classification
Relevant governance controls
Governance control mapping is not available for this record.
- No controls mapped
Not publicly disclosed
Control mapping is analytical. It does not state that any control would have prevented the incident.
Sources and evidence
This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.