Nadella calls for safeguards to contain AI models

Microsoft CEO Satya Nadella urged companies to design artificial intelligence systems so they can be observed and contained, saying they should assume models are compromised and ensure an authorized person can stop them while they are operating. In a post on X, Nadella argued that AI systems should not be treated as opaque tools whose recommendations and actions are simply accepted or rejected. He called for safeguards that make model behavior visible and leave tamperproof, human-readable evidence of meaningful actions. Nadella also said a single model should not control both a system’s behavior and the evidence used to verify it. That separation, he argued, would prevent an AI system from being responsible for its own independent check. He said companies should be able to reproduce how an outcome was reached without relying on the model’s own account. For systems with access to sensitive company data or the ability to take consequential actions, Nadella said organizations should apply an insider-risk approach. He argued that intelligence should be separated from authority, so a model may provide recommendations without being allowed to act unchecked.