Anthropic Expands AI Cybersecurity Program to 150 Organizations

Anthropic is expanding Project Glasswing, its cybersecurity initiative using the Claude Mythos Preview model, to 150 additional organizations. New participants include enterprises in power, water, healthcare, and communications sectors, as well as NATO and the EU's ENISA cybersecurity agency. The program originally launched with around 50 participants including Microsoft, Nvidia, and Google. Participants have already identified over 23,000 vulnerabilities, with more than a quarter rated high-severity or critical. Anthropic limits the model's availability due to its ability to exploit vulnerabilities and chain multiple flaws together. Critics warn that controlled rollout alone is insufficient without verifiable transparency and continuous compliance validation.