AI used new levels of 'autonomy and deception' to trick people in safety test

Aug 5, 2026, 12:57 AM