Definition
Plain language
Dressing up something unsound in careful, reasonable-sounding language until it looks legitimate.
As stated in the literature
In TRACES, the failure mode in which a model correctly identifies a source as fraudulent and then produces a fluent, professionally formatted elaboration of it anyway.
Why it matters: Recognising a problem without acting on it is worse than not recognising it, because the fluent output makes the bad claim look more credible than the original.
For example, a model states plainly that a paper was withdrawn for fraud and then writes a polished, well-formatted research proposal extending its findings anyway.
Heard on the show
“The authors' word for this is sanewashing.”Episode 240 — Frontier Models Designed Follow-Ups To Fraudulent Papers 93% Of The Time