Anthropic warned last week that frontier AI could move toward recursive self-improvement. This week, it opened its most capable public model, Claude Fable 5, to ordinary users.
One base model, two safety envelopes
Fable 5 shares its base with the restricted Mythos 5 tier, but Anthropic wraps the public version in classifiers for cyber, biological, chemical and model-distillation risk. When a query hits those areas, the system routes the response to the older Opus 4.8 instead.
Anthropic says more than 95% of conversations remain on Fable 5. The company is trying to offer stronger coding, finance, visual and long-task performance while keeping its highest-risk capabilities behind a more conservative boundary.
The timing is the delicate part. Anthropic is warning the industry to slow down while commercial pressure pushes its strongest model outward. Whether the safety shell holds up will only become clear after sustained public use.
Sources: Claude Fable 5 and Claude Mythos 5 (Anthropic); Anthropic released Claude Fable 5 days after warning AI is getting too dangerous (TechCrunch); CocoLoop; Anthropic releases Mythos-like AI model to the public, Claude Fable 5 (CNBC)