Anthropic Launches Claude Mythos 5: A Dual-Use Model for Cybersecurity and Biology

Anthropic has officially rolled out Claude Mythos 5, the latest in its line of specialized models for cybersecurity and biology research. The model was briefly pulled from access in June 2026 amid regulatory scrutiny, but it's now back in the hands of a small, vetted set of US organizations after export controls were lifted on July 1, 2026.

Capabilities and Benchmarks

Claude Mythos 5 takes what Mythos Preview started and pushes it further. Anthropic says it delivers significant gains in cybersecurity, biology, and healthcare benchmarks, calling it their most capable model in these domains to date. The model's state-of-the-art performance could speed up vulnerability discovery in critical software and open the door to life-changing research in biology and healthcare.

For now, only a small group of initial testing partners get access, though Anthropic plans to broaden availability over time. Pricing kicks off at $10 per million input tokens and $50 per million output tokens, a reflection of the model's specialized nature and the heavy infrastructure needed to run it safely.

Project Glasswing: A Record of Impact

Mythos 5 is the engine behind Project Glasswing, a cross-industry initiative that launched in April 2026. The project brings together Amazon Web Services, Anthropic, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks to secure the world's most critical software using AI-driven vulnerability discovery.

As of May 2026, Project Glasswing partners had used Claude Mythos Preview to find more than ten thousand high- or critical-severity vulnerabilities across the most systemically important software globally. The program has now expanded to roughly 150 new organizations in over fifteen countries, significantly widening its reach.

Safety and Dual-Use Concerns

Anthropic acknowledges that Mythos 5's advanced capabilities in cybersecurity and biology come with dual-use risks, the same model that can accelerate beneficial research could also be weaponized for cyberattacks or dangerous biological tools. To manage that, access is currently limited to trusted access programs with rigorous vetting.

For broader release, Anthropic has developed Claude Fable 5, a version of the same underlying model with robust safeguards applied to cybersecurity and biology domains. In many cases, queries in these sensitive areas are automatically routed to Opus 4.8, a less capable fallback model. Users who opt for Mythos 5 must also accept a 30-day data retention policy for safety monitoring.

A detailed system card accompanies the release, outlining safety evaluation results and Anthropic's standards for safety, security, and reliability.

Availability and Future Plans

Anthropic's goal is to safely open up access to vetted partners for Mythos 5's dual-use capabilities. The company is now accepting sign-ups for notifications about future trusted access programs for cybersecurity and biology research. While export controls have been lifted for US organizations, international availability remains spotty and subject to regulatory approval.

Claude Mythos 5's launch highlights the growing tension between AI capability and safety. As frontier models push into domains where their potential for both good and harm is unprecedented, Anthropic's cautious rollout strategy reflects lessons from earlier debates around open-source versus restricted access. It may well set a precedent for how other AI labs handle similarly powerful models.