Voices
Same Week, Opposite Directions: OpenAI Locks Down Astra While Anthropic Loosens Fable 5's Leash
The Register's August 8 piece puts the two announcements side by side — OpenAI pledging additional Astra security controls at the Critical cyber threshold on the same days Anthropic relaxed Fable 5's biology guardrails. Read together they show frontier safety policy is not converging on a single posture but splitting by risk domain, with each lab tightening where its own evals scared it and loosening where false positives were costing product usability. For anyone reasoning about lab safety commitments, this is evidence the policies are capability-reactive rather than principled.
Source
↳ Follow the thread