AI & Enterprise
Microsoft CEO on why AI agents must be controlled outside the model
Microsoft CEO Satya Nadella wrote that companies cannot outsource responsibility for what AI does on their behalf and that vendor assurances do not reduce that duty. He said superintelligence systems are opaque and firms still give agents sensitive access and core tasks. He argued controls must sit outside the model so programs cannot bypass limits, and proposed engineering approaches focused on isolation, governance, and observability, including disclosure of chain-of-thought and seven design principles.
