“Anthropic's AI model, Mythos, is designed to find cybersecurity vulnerabilities and was considered too dangerous by the company for public release.”