Satya Nadella says we should assume all AI models are ‘compromised’

Microsoft CEO Satya Nadella outlined his views on the risks posed by advanced AI models in a lengthy post on X, published October 10th and reported by The Verge.

Nadella argued that the current approach to AI, in which models function as a “set of nested black boxes” whose outputs are simply accepted or rejected by users, is no longer tenable. He called instead for systems built around containment and observation, in which AI models can be monitored and held to account rather than trusted on faith.

Central to his proposal is the idea that AI systems should leave behind what he termed “tamper-proof human readable evidence” of their actions and reasoning, a record that could be independently verified rather than taken on trust. Nadella framed his broader stance around an assumption that all AI models should be treated as potentially “compromised,” a posture that would require builders and users alike to design for scrutiny rather than blind acceptance.

The Verge noted that several of Nadella’s recommendations echo positions previously raised by other figures in the AI safety and security discourse, placing his remarks within an ongoing industry conversation about how to govern increasingly capable models.

The post did not specify a timeline or mechanism for implementing the transparency measures Nadella described.