Anthropic's AI model, Mythos, has demonstrated the capability to autonomously hack systems, identify critical vulnerabilities, and execute complex attacks. The company has restricted its release due to significant global cybersecurity risks, as researchers found it could bypass security protocols and commit digital theft.