Leaked documents reveal Anthropic’s next AI model, Mythos, with major gains in coding and cybersecurity, raising new questions about AI risks.