Anthropic Deploys "Claude Mythos Preview" for Cybersecurity, Citing Hacking Dangers

Anthropic has taken the unprecedented step of developing and selectively deploying a new AI model, 'Claude Mythos Preview,' which the company acknowledges possesses 'superhuman' capabilities in cybersecurity and hacking. Citing significant dangers of misuse if released publicly, Anthropic is instead rolling it out under 'Project Glasswing' to a tightly controlled group of 50-60 major technology and enterprise organizations, including Apple and Microsoft. The core strategy is to use this powerful offensive tool for defensive purposes, allowing these partners to proactively find and patch critical vulnerabilities in their software infrastructure before malicious actors can exploit them.

To facilitate this defensive arms race, Anthropic is committing substantial resources, including up to $100 million in AI usage credits for the participating organizations. This initiative represents a major shift in how frontier AI labs are handling dual-use technology, opting for controlled, private deployment over open release. The move highlights growing industry and regulatory concerns about the weaponization of advanced AI, setting a potential precedent for how other companies might handle similarly powerful and risky models in the future.