Nadella: AI Models Should Be Assumed Compromised From the Start

Microsoft CEO Satya Nadella said frontier AI models should be treated as insiders with access to critical systems. He outlined observability principles: containment, the ability for humans to pause a model mid-task, and tamper-proof audit trails.

• Nadella says models should be assumed compromised from the start
• An authorized person must be able to pause a model mid-task
• Using AI to verify AI creates nested black boxes
• Every model action must leave tamper-proof human-readable evidence