“Without safeguards, Fable 5's capabilities in areas like cybersecurity could be misused to cause serious damage.”

AnthropicAn AI safety company — creators of the Claude family of language models. split Fable 5 from Mythos 5 on day one: same weights, but Fable routes sensitive cyber and biology queries to Opus 4.8. The line justified why a frontier model shipped with guardrails critics called overly broad — and why Mythos stayed locked to Project Glasswing partners.

AnthropicAn AI safety company — creators of the Claude family of language models.'s own words for why its strongest GA model needed hard blocks on day one.

It framed dual-use AI risk as a shipping constraint, not just a research talking point.