Anthropic locked down its most powerful AI Model over cybersecurity fears–then put it to work

“Anthropic developed Claude Mythos Preview, its most capable AI model, to identify cybersecurity vulnerabilities across major operating systems and web browsers. Rather than releasing it publicly, the company discreetly deployed it through Project Glasswing, sharing findings with organizations responsible for internet infrastructure. This approach demonstrates how advanced AI can enhance cybersecurity while mitigating risks associated with releasing powerful models.”