Microsoft CEO Satya Nadella called for an "emergency brake" allowing authorized users to shut down artificial intelligence models mid-task, according to a post on X published Oct. 10, 2026.
Nadella argued that "Super Intelligence" systems should be treated as insider risks, saying the industry must separate the "supply of intelligence from the authority over it" to prevent responsibility for AI actions from being outsourced to model providers. He proposed surrounding non-deterministic models with deterministic system design and human controls, specifically separating the AI model from the "harness" that orchestrates its work and the action space that defines its capabilities.
Nadella outlined several principles for a new "trust architecture," including observability and evidence, model diversity, independent controls, and containment. He said every meaningful model action must leave "tamper-proof human readable evidence," and that no single model should be the sole dependency for an important outcome. He added that system design should assume a model is compromised from the start, requiring the ability for an authorized person to pause or shut down a model mid-task.
Nadella described Chain-of-Thought transparency as "non-negotiable," arguing that "Neuralese" should not justify opaque model reasoning, though he noted such transparency is insufficient because it is not yet known how to make model outputs consistently faithful. He recommended using models to adversarially test one another but warned this could create "nested black boxes."
Nadella concluded that the most trustworthy system is not one with the most trusted model, but one that "enables us to trust the model the least."