SOC’s new AI agents are risky: hallucinations, autonomy, and compromise threaten security operations.
