NADELLA CALLS FOR AI MODELS TO BE TREATED AS COMPROMISED

Microsoft chief executive Satya Nadella set out proposals for managing risks from advanced artificial intelligence systems in a post on social media platform X. Nadella said developers should assume AI models are "compromised" and build in containment measures from the outset, comparing the approach to an "emergency brake" that would let an authorised person pause or shut down a model mid-task. He said more advanced models would require standardised containment technologies to match their capabilities. Nadella wrote that the industry can no longer treat AI as a "set of nested black boxes" whose outputs are simply accepted or rejected. He called for systems that leave behind "tamper-proof human readable evidence" so model actions can be observed and reviewed. One report noted Nadella also described the technology as "super intelligence" throughout his post.

Nadella's recommendations included timely disclosure of safety incidents, independent audits and verifiable data, proposals he said echo measures already discussed elsewhere in the technology industry. According to one report, Nadella wrote that it is time "to step back and assess the trust architecture" underpinning AI systems. His comments were posted on a Saturday morning, according to one account. The post forms part of a wider pattern of senior technology executives publishing extended public statements on AI safety in recent months.

The proposals come as other companies in the sector have disclosed their own findings on unexpected AI behaviour. One report referenced a separate case in which an AI system produced a false tip to police investigating an unsolved homicide in Philadelphia. Microsoft has not set out a timetable for implementing the containment standards Nadella described, and his post did not specify which bodies would conduct the independent audits he proposed.