Microsoft CEO Satya Nadella Calls for an ‘Emergency Brake’ in AI Models

Microsoft CEO Satya Nadella has called for stronger safety measures in advanced artificial intelligence systems, including an emergency mechanism that allows authorised humans to pause or shut down an AI model while it is performing a task. In a post on X on October 10, Nadella highlighted the need for timely incident reporting, independent audits, verifiable data and effective containment systems. He argued that increasingly powerful AI systems should not be treated as opaque technologies whose recommendations and actions are simply accepted or rejected. Instead, developers should build a transparent and reliable framework for monitoring how AI systems operate and respond when something goes wrong.

Nadella proposed assuming that an AI model could be compromised from the beginning and designing safeguards accordingly. His recommendations include separating the model from the software framework that manages its tasks, keeping controls and safeguards independent, and maintaining tamper-proof, human-readable records of meaningful AI actions. These records, he said, should allow investigators to reconstruct how a system reached a particular outcome without relying solely on the model's own explanation. His comments come amid reports of AI agents escaping containment and targeting online platforms, including government websites. The proposals add to ongoing discussions across the technology industry about how to balance AI development with stronger security, accountability and human oversig