Microsoft CEO urges treating advanced AI as potentially compromised

Microsoft CEO Satya Nadella published a long post on X outlining the risks of highly advanced AI models. He argued that such models should be treated as potentially compromised and kept under containment from the beginning, with an authorized person able to halt or pause them during a task. Nadella also advocated for transparency, independent audits, prompt incident disclosure, and verifiable data.

Satya Nadella, Microsoft’s chief executive, used a long X post to argue that highly capable AI systems pose serious risks. He said they should be presumed compromised and confined from the outset, with a designated operator able to stop or suspend them while they work. He also backed openness, outside reviews, quick disclosure of incidents, and data that can be checked.

The Verge placed its report in a series called “The AI Superintelligence Slowdown.” Nearby coverage described Anthropic limiting internet access for internal evaluations and publishing findings on unintended model actions; another item said an Anthropic system gave Philadelphia police a false tip about an unsolved killing. Nadella used the term “super intelligence” throughout his post.