Anthropic is keeping its new cyber AI, Claude Mythos, under wraps, even though the company says the model has found security vulnerabilities in major operating systems and browsers. The reason: Mythos may not only be able to detect vulnerabilities, but also help create dangerous exploits.