Anthropic launches Claude Mythos 5 for cybersecurity, Fable 5 for public
In brief
- Claude Mythos 5 restricted to approved cybersecurity orgs, critical infrastructure operators, and government partners
- Claude Fable 5 delivers Mythos-class capabilities to general public with enhanced safety guardrails
- Mythos preview completed 32-step network intrusion exercise and discovered 271 Firefox vulnerabilities
Mythos 5: Restricted Access, Significant Capability
Anthropic announced the official launch of Claude Mythos 5, a restricted-access model representing a significant escalation in AI capability for offensive and defensive cybersecurity work. The company says the Mythos-class AI models "have reached a threshold where they present significant risks."
In April, Anthropic began Project Glasswing, releasing the first Mythos-class model to a limited group of cyber defenders and critical software infrastructure providers. The results were striking. The U.K.'s AI Security Institute reported that a preview version of Mythos became the first AI model to complete a 32-step corporate network intrusion exercise without human assistance. Mozilla reported that a preview version of Mythos discovered over 271 vulnerabilities in the Firefox browser.
Mythos 5 achieved a 78% score on ExploitBench, which measures vulnerability discovery and exploitation, and an 88% score on the Terminal-Bench coding benchmark. These benchmarks underscore the model's dual-use potential—capabilities useful for defenders but also for attackers.
Fable 5: Power With Guardrails
Claude Fable 5's capabilities exceed those of any model Anthropic has ever made generally available. The model shows exceptional performance across software engineering, knowledge work, vision, and scientific research.
To mitigate dual-use risk, requests involving cybersecurity, biology, chemistry, or AI model replication are handled by Claude Opus 4.8 as a safeguard triggered in fewer than 5% of sessions. This routing mechanism allows Fable 5 to deliver general-purpose capability while constraining high-risk use cases.
Expansion Plans and Government Alignment
Anthropic plans to steadily expand access to Claude Mythos 5 through a broader trusted access program in consultation with the U.S. government. The rollout reflects a careful balance between enabling cybersecurity defense and limiting offensive proliferation. Press reports have indicated that Anthropic had embedded engineers at the National Security Agency to help deploy Mythos for cybersecurity operations targeting China and Iran.
