Advanced artificial intelligence models should be treated as compromised from the start, requiring human operators to have an "emergency brake" that can shut them down at any moment, according to Microsoft CEO Satya Nadella. Writing in a detailed post on X, Nadella warned that the tech industry must rethink how it manages increasingly powerful automated systems before they cause harm.
Nadella argued that society can no longer accept AI as a "set of nested black boxes"—complex systems whose internal reasoning is hidden and whose actions people simply accept or reject without understanding. Instead, he called for transparent architectures where models can be closely monitored, contained, and required to leave behind "tamper-proof human-readable evidence" of their work.
While several of his recommendations align with broader industry discussions—including independent audits, verifiable data, and the prompt reporting of safety incidents—Nadella went further on containment. He argued that developers must assume every model carries hidden risks from the day it is deployed. Under his proposal, an authorized human supervisor must always retain the power to pause or shut down an AI model in the middle of a task.
As tech companies work toward more capable systems—which Nadella referred to throughout his post as "super intelligence"—he stressed that containment methods must be standardized across the field. In his view, stronger technical safeguards and clear human controls are essential to keeping advanced models under supervision.