Microsoft CEO Satya Nadella has shifted his public stance on AI, arguing that models should be treated as 'compromised' by default and must include a human-controlled 'emergency brake' to pause tasks mid-execution.