Microsoft CEO calls for ‘emergency brake’ controls on AI models

Microsoft CEO Satya Nadella argued that AI systems need a human-controllable “emergency brake” to pause or shut down models mid-task, framing the issue as a need to overhaul the “trust architecture” of AI rather than add isolated safety features.

Nadella’s post on X outlined three interlocking requirements: separating the model from the harness that orchestrates its work, externalizing controls and safeguards, and ensuring every meaningful model action is documented with tamper‑proof, human‑readable evidence.

He emphasized that an authorized person must always retain the ability to pause or shut down a model mid‑task, stating that systems should assume a model is compromised from the start and be designed to contain it.

The remarks follow a series of incidents where leading AI companies reported difficulties maintaining control over their models, and come after Anthropic CEO Dario Amodei published a plan advocating more cautious AI development.