Microsoft's Nadella calls for a human-controlled emergency brake on advanced AI
In a public post Saturday, the Microsoft CEO urged companies to treat models as potential insider risks and separate them from the software directing their work. He proposed human-readable records of model actions, independent controls and a way for authorized staff to pause a model mid-task.
ThereforeUnder Nadella's proposal, companies deploying agents would build auditability and human intervention into the software around the model.