Microsoft CEO Satya Nadella has argued that the tech industry must operate under the assumption that all advanced AI models are compromised. In a post on X, Nadella outlined a framework for addressing the risks associated with highly capable systems, asserting that the current approach to AI safety is insufficient.
Nadella rejected the prevailing model where AI functions as a "set of nested black boxes" whose outputs are simply accepted or rejected by users. He contended that this lack of transparency prevents proper oversight and risk management. Instead, he called for a system where models can be contained, observed, and where they leave behind "tamper-proof human readable evidence" of their actions.
The recommendations align with broader industry calls for timely incident disclosure, independent audits, and verifiable data. While these measures are widely discussed in safety circles, Nadella emphasized containment as a critical component. His post suggests a shift from trusting opaque systems to demanding auditable, transparent infrastructure for future AI deployments.