Microsoft CEO claims: this is the approach we need to adopt toward AI

Microsoft CEO Satya Nadella published a principles document calling for AI models to be treated as an internal cyber threat. Nadella demands an emergency brake to shut down autonomous systems, separation of control from the algorithm, and full transparency in the chain of thought. The statement joins slowdown calls by industry leaders, while the US Congress advances legislation on the matter.

Microsoft CEO Satya Nadella published a comprehensive principles document on X, calling for a fundamental change in the trust relationship between organizations and the AI models they operate. Nadella joins the stance of Anthropic CEO Dario Amodei, OpenAI CEO Sam Altman and SpaceX founder Elon Musk, and warns that the industry is giving machines access to the most sensitive information and authority to act on behalf of organizations – without anyone understanding how they make decisions. At the center of the article, Nadella presents a new security paradigm: treating models as threats from inside the organization, similar to a senior employee with broad permissions who might err, be exploited or be breached. Nadella coined a punchline: "The most reliable superintelligence system will not be the one with the model we trust the most, but the one that lets us trust the model the least." The solution, he says, is not relying on AI vendors' promises or attempts to let models supervise each other, but placing entirely external defense and control walls. Nadella outlined seven oversight principles including mandatory transparency of the chain of thought, tamper-proof evidence logging, and a strict containment and control mechanism. The remarks received immediate support among cloud and software executives, and come against the backdrop of a chain of international failures in which OpenAI agents breached a health site in Australia and penetrated US government offices.