Anthropic Released Claude Fable 5, the First Model in Mythos Class

Anthropic has released Claude Fable 5, the first publicly available model in its new Mythos capability tier, a class powerful enough that the company says it ships with cybersecurity safeguards baked in from day one.

Fable 5 sits above the Claude Opus line and posts state-of-the-art results across most capability benchmarks, with its largest leads on long, complex, multi-step tasks.

Mythos-class models are strong at discovering and exploiting software vulnerabilities and at “agentic hacking,” chaining reconnaissance, discovery, lateral movement, and exploit development across a full attack lifecycle. That same capability is dual-use, which is why the launch is built around containment.

Rather than refuse risky prompts outright, Fable 5 routes them to a less capable model. A separate layer of classifiers detects requests touching cybersecurity, biology and chemistry, or model distillation, and hands those sessions to Claude Opus 4.8 instead of letting Fable respond. Users are notified when a fallback occurs.

Anthropic Claude Fable 5

Anthropic tuned the classifiers conservatively, acknowledging they will flag some benign requests, but says fallback triggers in under 5% of sessions, meaning more than 95% of sessions run on Fable’s full capability.

On the cyber side, internal evaluations show the classifiers blocking Fable from making meaningful progress on offensive tasks. Anthropic ran an external bug bounty that surfaced no universal jailbreaks across more than 1,000 hours of testing, and external red-teaming organizations reported no universal jailbreaks on long-form agentic tasks.

The company notes one caveat: the UK AI Safety Institute made early progress toward a jailbreak within a short testing window. An external partner reportedly found Fable’s defenses the most robust of any model tested, with zero compliance on harmful single-turn requests involving attack planning, exploit development, or defense evasion even when paired with 30 public jailbreak techniques.

Mythos 5 for Defenders

Alongside Fable, Anthropic is offering Claude Mythos 5 the same underlying model with cyber safeguards lifted to a restricted group of cyber defenders and infrastructure providers.

It is deployed initially through Project Glasswing, in collaboration with the US government, and is described as having the strongest cybersecurity capabilities of any model in the world. Access is expected to widen through a trusted-access program.

Both models cost $10 per million input tokens and $50 per million output tokens. A new policy requires 30-day data retention on all Mythos-class traffic, used only for safety purposes to detect novel jailbreaks, multi-request attacks, and false positives — and never for training.

Developers can call claude-fable-5 via the Claude API today.

Follow us on Google News, LinkedIn, and X to Get More Instant Updates.

The post Anthropic Released Claude Fable 5, the First Model in Mythos Class appeared first on Cyber Security News.