Nadella: Treat Frontier AI Models as Insider Risk, Build ‘Emergency Brake’

Nadella's framing shifts the security burden from model developers to deployers, a notable stance for a CEO whose company both builds and deploys frontier models. Treating models as insider threats implies a governance gap: assurances from labs are treated as insufficient, and control layers must be owned by the operator. The practical question is whether deployers, many without that security depth, can build containment and kill-switch systems before agentic models are widely embedded in production workflows.