The weird situation with Fable

The video provides a detailed breakdown of the release of Anthropic's new 'Fable' model, a variant within the 'Mythos' class of models. The speaker argues that while the model shows promising performance, Anthropic has implemented highly restrictive, 'unacceptable' safety measures that include silently rerouting user requests to a fallback model (Opus 4.8) when specific keywords are triggered. This behavior is problematic because users are often unaware they are being rerouted, and it creates a supply-chain risk for developers who build dependencies on the model's behavior. The speaker also highlights a lack of transparency regarding changes to technical documentation, with evidence of Anthropic modifying their system cards to conceal specific safety interventions. Ultimately, the video argues that these actions undermine user trust and set a dangerous precedent by allowing AI providers to alter model performance unilaterally and without adequate disclosure.