Anthropic Splits Claude Access Between Specialized Research and Public Use

Anthropic is bifurcating its model deployment strategy to manage the inherent risks of high-capability AI. The company has officially launched Claude Mythos 5, a restricted-access model for cybersecurity and scientific research, alongside Claude Fable 5, a version intended for broader public use [Anthropic rolls out Claude Mythos 5 AI model—along with the safer Fable 5 for the public].

The deployment of Mythos 5 is not universal. Access is limited to approved cybersecurity organizations, government partners, critical infrastructure operators, and selected life sciences researchers. This restriction follows Anthropic's admission that Mythos-class models have reached a threshold where they present significant risks.

The technical capabilities of the Mythos lineage have already demonstrated autonomous offensive potential. The U.K.'s AI Security Institute reported that a preview version completed a 32-step corporate network intrusion exercise without human assistance. Furthermore, a report from Mozilla indicated that a preview version of Mythos discovered over 271 vulnerabilities in the Firefox browser.

This dual-track release—Mythos 5 for specialized sectors and Fable 5 for the public—is an attempt to scale capability while implementing safety guardrails. Fable 5 is designed to route some sensitive requests to a less capable model to prevent misuse.

The move signals a shift in how frontier labs must handle models that can identify software vulnerabilities and conduct complex cybersecurity operations. As these models move from experimental previews to official launches, the boundary between research utility and autonomous threat becomes the primary friction point for regulators and developers alike.

The central question for the industry is whether a "safer" version of a high-capability model like Fable 5 can truly mitigate the risks identified in the Mythos-class architecture.