An inexperienced hacker managed to compromise over a dozen companies using AI agents to do most of the work, raising real questions about AI safety.