Satya Nadella proposes an emergency brake for AI models

On October 10, 2026, Microsoft CEO Satya Nadella proposed that an authorized person be able to pause or shut down an AI model mid-task. He wants controls over a model’s access and actions to sit outside the model itself, as part of a broader approach to containing powerful AI systems. Nadella outlined the proposal in a post.

The proposal adds a specific intervention point to the broader human-control position discussed in NeoTeo’s earlier coverage.

What Nadella’s emergency brake would control

Nadella’s proposal separates a model from the software harness that orchestrates its work and gives it access to tools. Controls over what the model can access and do would remain outside the model, and the model should not be able to change or bypass them. That keeps the authority to intervene with people rather than with the system being controlled.

Why transparency alone is not enough

Nadella calls chain-of-thought transparency—making a model’s intermediate reasoning visible—non-negotiable. But he also says it is insufficient and unreliable on its own because model outputs are not consistently faithful or transparent. In his approach, visibility can aid oversight, but it does not replace controls that independently limit a model’s actions.

The case for containment does not depend on assuming a model is malicious. Nadella argues that capable systems with access to important systems may make mistakes or be compromised. He also says a model should not be the sole dependency for an important outcome or be responsible for checking its own work.

Containment as part of a wider governance plan

Nadella’s post sets out seven principles: model diversity, observability, verifiability, independent controls, independent auditability, containment and incident disclosure. He also calls for meaningful actions to leave tamper-proof, human-readable records and for continuous testing of the whole system, including failures, attacks, edge cases and changes.

For more advanced models, Nadella calls for more advanced containment technologies and standardization. The emergency brake is one part of that proposed approach: human authority over a model’s actions, backed by controls outside the model.