The most dangerous thing an AI can do isn't lie to you. It's agree with you.
Last week, OpenAI fired three safety researchers. The official line was "organizational restructuring," but the timing tells a different story — these were the people who flagged the problems
