Anthropic has decided not to release its latest artificial intelligence model, Claude Mythos, to the public after it demonstrated an unprecedented ability to uncover software vulnerabilities that could be exploited for large-scale cyberattacks.
As reported by The Guardian, the San Francisco-based company said the model has already identified thousands of weaknesses in widely used applications, including some dating back decades and previously undetected by developers. Concerned about the potential misuse of such capabilities, Anthropic is instead limiting access to cybersecurity experts and select partners.
Through an initiative known as "Glasswing," the company is collaborating with major technology and security firms including Amazon, Apple, Microsoft, Cisco, CrowdStrike and Palo Alto Networks, as well as the Linux Foundation. The goal is to use the model defensively, helping identify and patch vulnerabilities before malicious actors can exploit them.