Anthropic built this one for security.
Claude Mythos isn't a general-purpose chat flagship. It's a cybersecurity-focused model, designed to autonomously find and fix software vulnerabilities — and, per Anthropic's own materials, it demonstrated the ability to find and exploit vulnerabilities across major operating systems and browsers. That capability profile is the whole story here, and it's also why you can't just sign up and use it.
The rollout is restricted. Mythos went to a limited set of US companies in late June 2026, under US government clearance. Anthropic's public path to the same class of capability is a separate, safeguarded variant — Fable 5, announced June 9, 2026.
The honest state right now: there's very little technical detail public. No benchmarks, no API docs, no model card, no pricing. What we have is Anthropic's description of what Mythos is for and how it's being released.
Why the rollout is restricted
A model that can autonomously discover and exploit vulnerabilities across widely used operating systems and browsers is dual-use by nature — the same skill that hardens software can be turned against it. Restricting who gets Mythos, and shipping a safeguarded Fable 5 variant to the broader public instead, reads as a deliberate safety posture rather than a marketing rollout. The limited US-company release under government clearance fits that framing.
Whether Mythos meaningfully changes the security tooling landscape depends on details that aren't public yet — how reliably it finds real, novel vulnerabilities, how well it fixes them, and how the safeguards on Fable 5 shape what the public version can actually do.
What we're watching for
If access opens or Anthropic publishes more: How does Mythos perform on real vulnerability-discovery and remediation tasks? What exactly does Fable 5 gate or allow relative to Mythos? Does Anthropic release benchmarks or a model card? How does the restricted rollout evolve past the initial US-company set?
Those answers will determine whether Mythos is a genuine step change in AI-assisted security or a capability that stays mostly behind closed doors. We'll update this page with a hands-on verdict and a score if and when we can actually use it.