Microsoft CEO Satya Nadella has called for a security posture that assumes advanced AI models may already be compromised and keeps them contained from the outset. According to The Verge, which reported on the post on October 10, 2026, Nadella set out the position in a lengthy post on X.
What Nadella is proposing
Per The Verge, Nadella argues that highly advanced AI models carry dangers that must be confronted. He says society can no longer accept AI as nested black boxes whose outputs are simply taken or rejected.
In their place, he describes a more transparent system built around three properties:
- Models can be contained.
- Models can be observed while they run.
- The system leaves tamper-proof, human-readable evidence of what happened.
The central security principle, as reported, is that a model should be assumed compromised and contained from the start. That reverses the usual order, in which trust is extended first and revoked when something goes wrong.
An "emergency brake" and standard containment
Nadella also says an authorized person should always be able to pause or shut down a model in the middle of a task. The Verge calls this an "emergency brake."
He adds that more capable models will need more capable containment technologies, and that these should be standardized.
How it compares with other industry calls
The Verge says Nadella's recommendations overlap with those of other industry figures: timely disclosure of incidents, independent audits, verifiable data and containment. In the outlet's assessment, his stance on containment goes somewhat further than some others in the field.
The Verge also notes that Nadella uses the term "super intelligence" for AI throughout the post.
