Satya Nadella Says AI Models Should Be Assumed Compromised

Summary

Microsoft CEO Satya Nadella has outlined his concerns about the risks posed by highly advanced AI models and proposed a more controlled approach to operating them. In a lengthy post on X, he argued that society should move away from treating AI as a set of nested black boxes whose advice and actions are simply accepted or rejected. Nadella called for systems that can be contained and observed while producing tamper-proof, human-readable evidence of what they do. His recommendations also include timely incident disclosure, independent audits, and verifiable data. The strongest part of his proposal is that organizations should assume a model is compromised from the beginning and contain it accordingly. He compared this control to an emergency brake, saying an authorized person should always be able to pause or shut down a model in the middle of a task. Nadella added that more advanced models will require more advanced containment technologies and argued that these mechanisms should be standardized. The article notes that he repeatedly uses the term “super intelligence” when discussing advanced AI.