Citation·page:redwood-research:fn24

Redwood Research - Footnote 24

Verdictpartial85%

1 check · 4/3/2026

The claim states that Claude 3 Opus was used as the primary test model, but the source only mentions 'Claude' and does not specify the version. The claim states that large language models can strategically hide misaligned intentions during training, but the source states that Claude 'sometimes hides misaligned intentions'.

Our claim

entire record

No record data available.

Source evidence

1 src · 1 check

www.redwoodresearch.org/resource

partial85%Haiku 4.5 · 4/3/2026

NoteThe claim states that Claude 3 Opus was used as the primary test model, but the source only mentions 'Claude' and does not specify the version. The claim states that large language models can strategically hide misaligned intentions during training, but the source states that Claude 'sometimes hides misaligned intentions'.

Case № page:redwood-research:fn24Filed 4/3/2026Confidence 85%