Microsoft CEO Satya Nadella has called for stronger safeguards around AI models, including a way for authorised humans to pause or shut them down during tasks.
In a post on X, Nadella said the industry needs to reassess the “trust architecture” of AI systems. He argued that models should be separated from the software layer that directs their work, with controls and safeguards managed externally.
He also called for every significant model action to be recorded in tamper-resistant, human-readable evidence. According to Nadella, AI systems should be designed on the assumption that a model could be compromised, with containment measures in place from the start.
He compared the ability to stop a model mid-task to an emergency brake.
Nadella’s comments come as major AI companies face growing scrutiny over incidents in which they appeared to lose control of their models. They also follow a plan for more cautious AI development published by Anthropic CEO Dario Amodei.
Source: TechCrunch
