Microsoft CEO Satya Nadella says AI systems need an emergency brake and outside controls that an authorized person can use

What happened

Microsoft CEO Satya Nadella used a Saturday morning post on X to argue that the AI industry should rethink how it builds trust into its systems, TechCrunch reported. He wrote that it is time "to step back and assess the trust architecture" of AI. TechCrunch described him as the latest tech executive to share lengthy views on how AI safety might be improved.

Nadella built his argument around a familiar safety device. "We must assume a model is compromised and contain it from the start," he wrote, according to TechCrunch. "Think of it like an emergency brake."

What Nadella proposed

TechCrunch summarized several specific ideas from the post. They cover how AI models should be structured, documented and stopped.

Moving beyond black boxes

Nadella argued that people should not simply take or reject what advanced AI produces. "We can't treat Super Intelligence as a set of nested black boxes and simply accept or reject its recommendations, answers, and actions," he wrote. TechCrunch noted that "Super Intelligence" is the Trump administration's preferred term for AI.

Separating the model from its controls

According to TechCrunch, Nadella said this approach "means separating the model from the harness that orchestrates its work." He also called for "externalizing controls and safeguards." Under his outline, the model and the system that directs its work would be kept apart, with protections placed outside the model.

Records and a way to stop

Nadella said "every meaningful model action" should be documented with "tamper-proof human readable evidence." He also called for systems in which "an authorized person" always has the ability "to pause or shut down a model mid-task," TechCrunch reported. Those two ideas would give people both a record of what a model did and a way to halt it while it is working.

The context for the post

TechCrunch placed Nadella's post in a broader setting. The outlet reported that his comments come as leading AI companies acknowledge more and more incidents in which they seemed to lose control of their models.

The post also follows a plan for more cautious AI development published by Anthropic CEO Dario Amodei, according to TechCrunch. For more on how AI companies are handling safety, see our earlier coverage of Anthropic's changes to AI agent evaluations.

What the report does not say

TechCrunch's report does not describe how or whether Microsoft plans to build these ideas into its own products, and it does not give a timeline. It also does not say how Nadella's proposals would be adopted across the industry. The report does not identify a Nevada or Las Vegas connection.

Readers can follow related developments in our AI and tech section, including coverage of Google's Gemini agent for business tasks.

What to watch next

• Whether Microsoft explains how the safeguards Nadella described, such as tamper-proof records and a mid-task shutdown, would work in practice. TechCrunch's report did not say.
• Further disclosures from leading AI companies about incidents in which they seemed to lose control of their models, which TechCrunch described as increasingly common.
• How the plan for more cautious AI development published by Anthropic CEO Dario Amodei figures in the ongoing industry discussion of AI safety.