Anthropic's Claude Fable 5 delivers Mythos-class AI with safety guardrails for general users

Anthropic has unveiled Claude Fable 5, positioning it as a 'Mythos-class model made safe for general use.' Fable 5 shares the same base architecture as Mythos—a highly restricted model previously available only to Project Glasswing partners like AWS, Apple, Google, and Microsoft. Mythos itself is now rolling out more broadly as Mythos 5 under a trusted-access program. Fable 5, however, is designed for all users, with pricing around double that of Claude Opus 4.8. To maintain safety, Anthropic added classifiers that block responses in high-risk areas of cybersecurity and biology. When a prompt triggers these guardrails, the model automatically falls back to Opus 4.8, which has its own restrictions (e.g., blocking ransomware code development). This fallback ensures no dangerous outputs while preserving general utility.

Anthropic shared strong early safety data: 'At least 95% of Fable sessions run entirely on Fable's own responses, with no fallback.' The company ran an external bug bounty that produced no universal jailbreaks in over 1,000 hours of testing, and external red-teaming organizations also failed to find universal jailbreaks. Customer feedback highlights Fable's deep coding capabilities: an unnamed Base44 representative said Fable is 'much deeper and better at one-shotting full apps, and its tool calling is excellent.' For professionals, Fable 5 offers Mythos-level power for coding and tool use, but with guardrails that may frustrate security researchers. Those with Anthropic's Cyber Verification Program access may retain elevated permissions, though that's not yet confirmed.