Microsoft Calls for 'Emergency Brakes' to Control AI Models
Microsoft CEO Satya Nadella called for separating models from the systems managing them and for tamper-proof logs of their actions, while enabling any model to be stopped mid-task.
In a post on the X platform, Microsoft CEO Satya Nadella called for a reassessment of what he described as the trust architecture in artificial intelligence, as reported by TechCrunch. He stated that it is time to step back and review the way we build trust in these models.
Nadella called for decoupling AI models from the systems that manage and direct their operation, and for tamper-proof, human-readable logs of every material action performed by the model. He also demanded that an authorized person be able to stop or shut down any model mid-execution.
This call comes after multiple incidents where companies reported losing control over their models or agents, according to the platform. Nadella's proposals aim to provide operators with verifiable oversight tools, rather than relying solely on voluntary commitments.
This development affects everyone providing AI-based services, as the absence of a verifiable log makes determining accountability for any failure difficult. The statement also reflects a shift in the tone of major model developers toward safety rules at a time of growing regulatory demands on this issue.
What do these terms mean?
Trust architecture: The set of rules and safeguards that make it possible to rely on a system and verify its actions.
Tamper-proof log: A log that cannot be modified later and records the model's actions in a format that can be read and reviewed.
Emergency brake: A mechanism that allows an authorized person to immediately stop an automated system while it is operating.
Weekly Newsletter
Read between the lines before everyone else. Decode the most important economic, tech, and decision-maker movements in the region.. in 5 minutes every Saturday.
