Nadella says assume all AI models are compromised

microsoft ceo satya nadella advocates for a shift in how ai models are treated, suggesting they should be assumed compromised and equipped with emergency brakes. nadella proposes building transparent systems where models can be contained, observed, and leave behind tamper-proof evidence. he recommends timely incident disclosure, independent audits, and verifiable data. nadella's emphasis on containment, including the ability to pause or shut down models mid-task, aligns with but goes beyond some industry recommendations.