Microsoft CEO Satya Nadella Calls for ‘Emergency Brake’ on AI Models

Microsoft CEO Satya Nadella has called for a fundamental restructuring of how artificial intelligence models are built and monitored, emphasizing the need for robust safety mechanisms. In a post published Saturday morning on X, Nadella argued that the industry must pause to evaluate the “trust architecture” underpinning AI systems.

Nadella warned against treating advanced intelligence as a series of opaque, nested black boxes where users merely accept or reject automated outputs. He specifically used the term “Super Intelligence,” noting it as the preferred language of the current administration, to describe the technology requiring these safeguards.

The Microsoft leader outlined a framework that separates the core model from the “harness” that directs its operations. Key elements of this proposal include externalizing controls and safeguards, as well as ensuring that every significant action taken by a model is recorded with tamper-proof, human-readable evidence. Crucially, Nadella insisted that an authorized individual must always retain the ability to pause or terminate a model while it is performing a task.

“We must assume a model is compromised and contain it from the start,” Nadella wrote, comparing the necessary control measures to an emergency brake. His comments arrive amid growing concern within the tech sector over AI safety, following recent reports of leading companies struggling to maintain control over their systems. This follows a plan published last month by Anthropic CEO Dario Amodei advocating for a more cautious approach to frontier AI development.