Microsoft CEO outlines safeguards for increasingly capable AI
Microsoft CEO Satya Nadella says advanced AI systems need stronger safeguards, including a reliable way for people to stop them while they are carrying out tasks. He argues that AI should not be treated as a black box whose recommendations and actions are simply accepted or rejected.
Nadella's proposed approach separates an AI model from the systems that direct its work, while placing controls and protections outside the model itself. He also calls for meaningful actions to be recorded in tamper-resistant, human-readable logs.
A further safeguard would ensure an authorized person can pause or shut down a model mid-task. Nadella describes this capability as an "emergency brake" and says developers should plan for the possibility that a model's behavior or security has been compromised.
These ideas point toward AI systems designed with oversight and containment built in from the start. If adopted widely, such controls could help make increasingly capable AI more accountable and manageable.
