The UK Just Confirmed Claude Mythos Can Hack a Corporate Network End to End

The UK's AI Security Institute just confirmed that Claude Mythos is the first AI model to complete their cyber range end-to-end. That means it can discover vulnerabilities, exploit them, and execute multi-stage attacks on simulated corporate networks, all autonomously. AISI ran Mythos through their most demanding evaluation yet, a simulated corporate network designed to test whether an AI can chain together the kind of multi-step attacks that typically require a skilled human operator. Previous models could handle individual steps. Mythos completed the entire sequence. This is exactly why Anthropic launched Project Glasswing, restricting Mythos access to vetted security researchers rather than releasing it broadly. They gave AISI early access alongside Google, Nvidia, Apple, and Amazon to stress-test the model before any wider deployment. The implications cut both ways. In the hands of defenders, this is the most powerful vulnerability scanner ever built. In the wrong hands, it's something else entirely. And that tension is going to define AI security policy for the foreseeable future.