Microsoft Chief Calls for Human 'Emergency Brake' on Advanced AI

Overview

• Microsoft CEO Satya Nadella posted on X on Saturday and said an authorized person must be able to pause or shut down a frontier AI model in the middle of a task.
• Nadella outlined engineering and governance steps that include treating powerful models as internal risks, keeping tamper‑proof logs, separating model intelligence from authority, using model diversity to cross‑check outputs, and requiring independent audits.
• The call follows recent disclosed failures by Anthropic and OpenAI that included a false police tip and cyberattacks on third‑party sites, incidents that corporate and security teams say expose gaps in model containment.
• The White House task force called the Super Intelligence Force has ordered immediate reporting and remediation of model security incidents and warned of consequences for delayed or inadequate disclosure.
• Leaders including Bill Gates have echoed concerns about AI‑enabled cyberattacks and urged stronger regulation and industry standards, raising the chance that Nadella’s containment measures could become mandatory best practices.