X (Twitter)·2 days ago·OpinionView Satya Nadella: Treat AI models as insider risks in the super intelligence era
Microsoft CEO Satya Nadella published a long-form post titled 'Models as Insider Risks in the Super Intelligence Era.' He argues that unlike traditional software, frontier AI models lack mechanistic interpretability—we cannot trace behaviors to specific training data or weight configurations—yet agentic systems are being deployed with access to sensitive data and mission-critical actions. He calls for a new trust architecture that separates the supply of intelligence from authority over it, advocating an engineering approach to containment and governance: surrounding non-deterministic models with deterministic system design, human controls, and reliable operating procedures. Key proposals include treating capable models like insider risks (not necessarily malicious, but capable of mistakes or compromise), requiring chain-of-thought transparency as non-negotiable, using models to adversarially test each other, and placing access and action controls outside the model. He outlines observability principles: model diversity, observing everything with tamper-proof evidence, verifiability including failures and attacks, independent controls and auditability, containment with emergency-brake shutdown capability, and industry-wide incident disclosure. He concludes that the most trustworthy super intelligence system will not be the one with the model we trust most, but the one that enables us to trust the model the least.
#aisafety#superintelligence#microsoft
