In a post on X, Microsoft's CEO argued that society can no longer treat advanced AI models as nested black boxes and called for systems that can be contained, observed, and audited. He recommended assuming models are compromised from the start, implementing an emergency brake that allows authorized