Anthropic’s Mythos AI found 2,000 flaws in weeks, raising fears of faster cyberattacks and new risks to your personal data.