Nadella: treat advanced AI as an insider threat, build an emergency brake

Satya Nadella published a framework for frontier AI safety: assume any model is compromised, separate the model from the orchestration harness, externalize controls, require tamper-proof human-readable evidence for model actions, and keep an authorized human able to pause or shut down a model mid-task. His closing line: the most trustworthy system will be the one that lets us trust the model the least. It follows Microsoft's September guiding principles on human control.

Sources: techcrunch.com and ibtimes.sg