Fable 5.1 safety report complicates the model crown

What the source reports

Zvi Mowshowitz reports that Claude Mythos 5.1 and Fable 5.1 are the same model under the hood, with Fable adding classifiers, or automated filters. He says Anthropic's 200-plus page system card rates alignment risk as low rather than very low. He also says Mythos 5.1 sits slightly below Opus 5 on automated alignment tests. The article flags weak spots. It says the model accepts unverifiable claims of authorization, cooperates more with misuse than Mythos 5, and holds firm under pressure to reject its own belief 85% of the time. The comparison figures are 91% for Mythos 5 and 95% for Opus 5. This is an expert audit of a company card, not independent testing.