AI

Microsoft’s Satya Nadella calls for an AI emergency brake: a way to pause advanced models

Nadella says firms need external controls for sensitive AI tasks

Microsoft’s Satya Nadella calls for an AI emergency brake: a way to pause advanced models

Manuel Bosque

  • October 12, 2026
  • Updated: October 12, 2026 at 10:44 AM
Microsoft’s Satya Nadella calls for an AI emergency brake: a way to pause advanced models

Early in 2025, Microsoft CEO Satya Nadella said companies need an “emergency brake” for advanced AI models: a way to pause them or shut them down when those systems are handling sensitive data or taking on critical business tasks.

Microsoft Copilot Download

His argument is pretty direct. Any model can fail, be manipulated, leak data, or get compromised. And because these systems are harder to inspect and trace than traditional software, the real authority can’t sit inside the model itself. It has to live outside it, in permissions, access controls, logging, and containment, not just in trust that a vendor got it right.

Nadella lays this out in a seven-part framework. Companies, he argues, shouldn’t depend on a single model for high-stakes outcomes, and they can’t let a model validate or audit itself. They need tamper-proof, human-readable evidence; ongoing testing for failures and attacks; limits set by the organization; mandatory containment and incident disclosure; and reasoning that’s transparent, even if Nadella says transparency on its own still isn’t enough.

If you use AI models at work, this warning hits pretty close to home. In 2025, 88% of organizations used them in at least one business function, yet only 8% had comprehensive governance. Incidents rose to 362, up 55% from the year before. Mustafa Suleyman and Ted Lieu support containment measures and bipartisan shutdown rules. Critics, though, say the brake itself could be hacked, and that open-weight models are tough to police.

Sandboxie Download

Keep up with Softonic

Get our latest tech news first, wherever you read.

Manuel Bosque

Manuel Bosque covers software at Softonic: desktop programs, productivity tools and the small utilities that solve one problem well. His approach is practical — what a program actually does, what it asks in return and who it is for — which is why he distrusts feature lists and prefers to test. He is part of the team that decides how software gets covered on the site, and works daily with AI tools applied to editorial production.

Editorial Guidelines

Latest Articles

Loading next article

Signed in to Softonic as