Microsoft's Nadella on October 10 calls for an emergency brake on AI models and separation of model and harness

Microsoft CEO Satya Nadella posted on X on October 10 calling for a trust-architecture reset for the era of superintelligence. He argued that superintelligence cannot be treated as a stack of nested black boxes whose recommendations, answers, and actions are simply accepted or rejected. His proposal has five pillars: separate the model from the harness that orchestrates its work, externalize controls and safeguards, require tamper-proof human-readable evidence for every meaningful model action, allow any authorized person to pause or shut down a model mid-task, and assume from the start that a model is compromised and contain it. Nadella framed the package as an emergency brake, the first time a US frontier CEO has used the Trump administration's superintelligence vocabulary to argue for a structured AI safety framework. The post follows a string of model-control incidents disclosed by leading AI labs, including Anthropic's October 9 cut of internal agent evaluations from live internet access. The shared vocabulary signals an emerging alignment between the largest US tech companies and the federal AI policy lexicon in late 2026.