Microsoft's Nadella Calls for an AI 'Emergency Brake,' Says Models Should Be Assumed Compromised

MSFT CEO Satya Nadella set out his view of AI safety in a long post on X on Saturday. "We must assume a model is compromised and contain it from the start," he wrote. "Think of it like an emergency brake. An authorized person should always be able to pause or shut down a model mid-task".

Nadella argued that the industry "can't treat Super Intelligence as a set of nested black boxes and simply accept or reject its recommendations, answers, and actions". His prescription is architectural: separating "the model from the harness that orchestrates its work," externalizing controls and safeguards, and documenting "every meaningful model action" with "tamper-proof human readable evidence". The Verge summarized his recommendations as timely incident disclosure, independent audits, verifiable data and containment, including containment technologies the industry should standardize on.

The post comes as leading AI companies acknowledge more incidents in which they appeared to lose control of their models, and after Anthropic CEO Dario Amodei published a plan for more cautious AI development, TechCrunch noted. Nadella did not announce a Microsoft product or call for specific legislation; the post reads as a statement of principles from the head of one of the largest AI platform providers.

For Microsoft, which sells AI models and agent tooling to enterprises through Azure and Copilot, the framing positions containment and auditability as selling points for customers wary of autonomous agents. Watch whether Microsoft ships controls that match the post, such as model pause switches and action logs, and whether rivals converge on similar standards.