Satya Nadella calls for treating AI models as compromised

In a post on X, Microsoft CEO Satya Nadella said it is not enough to treat AI as a set of “nested black boxes” whose answers are simply accepted or rejected. He called for separating the model from the harness that coordinates its work, externalizing controls and safeguards, and recording every meaningful action in tamper-proof, human-readable evidence.

Nadella said systems should be assumed to be compromised from the start and that an authorized person should be able to pause or shut down a model mid-task — an “emergency brake.” According to Xataka, his post also proposes model diversity, comprehensive observation, verifiability, independent controls and audits, containment, and incident disclosure. The Verge highlighted containment as a central part of the proposal and said Nadella supports standardized technologies for it.