Anthropic is shipping two versions of its latest Claude model, and which one you get depends on how much the company trusts you.
Claude Fable 5 is the public release, built with constraints Anthropic says make it unusable for cyberattacks. Claude Mythos 5 is reserved for vetted "cyber partners," organizations that have cleared some unstated bar of responsible use. The exact criteria for that vetting are not publicly detailed, and neither is who reviews the list over time.
The two-tier structure reflects a genuine tension in AI deployment: the same capabilities that help security professionals with threat modeling and red-teaming can be turned toward offensive ends. Anthropic is betting that access control is a better solution than capability limits alone. The approach raises an obvious question: if Fable 5's guardrails actually prevent cyberattacks, what does Mythos 5 unlock that justifies a separate tier?
Tiered access is increasingly an industry pattern, not an Anthropic invention, but it shifts accountability questions off the model and onto the vetting process, which tends to receive less scrutiny.