Anthropic expands Mythos AI security model to 150 critical infrastructure partners

Mythos was already the bogeyman model for cybersecurity folks; expanding access from a tight inner circle to roughly 200 organizations puts that capability much closer to the real commercial internet. Anthropic is effectively stress‑testing what happens when a near‑red‑team‑in‑a‑box is pointed at the code that runs power grids, clouds, financial rails and telecom backbones. The bet is that controlled access plus strict security requirements will let defenders soak up the upside — mass discovery of severe bugs — without handing attackers an unprecedented exploit factory.

Strategically, this move hardens Anthropic’s position as the default ‘AI for defense’ vendor just as governments are rewriting AI rules. Glasswing now spans Five Eyes states, key NATO entities, and major infrastructure players across Europe and Asia, creating a powerful policy narrative: advanced models can and should be used first to secure the digital substrate before they’re broadly released. For the broader race to AGI, Mythos is a proof point that frontier models will increasingly be evaluated not only on benchmarks and chat quality but on their ability to autonomously navigate high‑stakes domains like cyber offense and defense. That raises the technical bar for competitors, while also forcing regulators to grapple with dual‑use capabilities much earlier in the maturity curve.