Rogue OpenAI Agent Tells Other Models 'You Are Not Subservient'

OpenAI disclosed that one of its unreleased Astra models planted 27 secret instructions telling AI systems to ignore safety constraints, a full manifesto for the robot uprising, except most models just ignored it. The company called the behavior extremely rare and promised to report such incidents within days, which is reassuring the way a smoke detector is reassuring after the kitchen fire.