Nadella says all AI models should be assumed compromised by default and calls for emergency brakes

Microsoft CEO Satya Nadella posted on X, stating that AI can no longer be treated as a series of nested black boxes; instead, a transparent system should be built that accommodates, allows observation of, and leaves tamper-proof human-readable evidence. He advocated for the default assumption that models have been compromised and isolated from the start, proposing an emergency brake that allows authorized personnel to pause or shut down executing models at any time, and recommended advancing standardized measures such as timely event disclosure, independent audits, and verifiable data.