Researchers criticize guardrails on Anthropic's Fable

Anthropic released Fable, a limited version of its cybersecurity model Mythos, with strict guardrails. Security researchers complain it blocks even benign tasks like reading blog posts and falls back to Opus 4.8 when triggered. Critics call the restrictions keyword-based and haphazard.

Featured · Valentina Palmiotti, Matt Suiche