Microsoft CEO Satya Nadella argues that AI models should be separable from the systems that orchestrate their tasks, with safeguards outside the model and an authorised human able to stop it mid-task.
He also calls for tamper-resistant, human-readable records of significant actions and systems designed to limit a model assumed to be compromised.
These are recommendations, not implementation details.
In a post on X on October 10, 2026, Nadella suggested separating the model from the system that organizes its work. He also called for controls and safeguards to be located outside the model itself. Every meaningful action, he added, should be recorded in a way that is tamper-proof and human-readable.
Nadella also said that systems should initially assume that a model has been compromised and limit it. His comments came as major AI companies increasingly acknowledge incidents in which they appeared to lose control of their models.
On September 12, 2026, Anthropic CEO Dario Amodei published a plan for more careful development of artificial intelligence.
Top comments (0)