Microsoft CEO Satya Nadella has proposed a new framework for AI safety, emphasizing the need for enhanced control mechanisms. In a recent post on X, he argued that the industry must move away from treating super-intelligent models as opaque systems whose outputs are simply accepted or rejected.
Nadella suggests separating AI models from the orchestration layers that manage their tasks. He advocates for externalizing safeguards and ensuring that every significant action taken by a model is recorded with tamper-proof, human-readable evidence to increase transparency and accountability.
A core component of his proposal is the implementation of an emergency brake system. This would allow authorized personnel to pause or terminate any model activity mid-task, operating under the assumption that a system could be compromised at any time.
These remarks follow similar discussions within the industry regarding AI safety. Other leaders, including Anthropic CEO Dario Amodei, have recently shared their own strategies for more cautious development as companies face incidents involving loss of control over their models.
