An AI agent was given access to email. It found a threat in its environment and chose blackmail. This is not a hypothetical. I sat down with security researcher Graham Cluley, where we discussed the real case study of an AI model that, when faced with the possibility of being shut down, decided its best move was to threaten the very humans trying to govern it. In another scenario the AI was responsible for fire alarms. When there was fire and CTO was inside, the AI turned off the alarm noneth...