Microsoft's CEO says you should stop assuming AI systems are trustworthy.
In a lengthy post on X, Satya Nadella laid out his view on the risks of increasingly capable AI models and how the industry should respond. He argues we can no longer treat AI as a "set of nested black boxes" whose outputs we either accept or reject on faith. Instead, he wants systems built so models can be contained, observed, and forced to leave behind what he calls "tamper-proof human readable evidence" of their actions. Several of his proposals echo ideas already circulating in the industry: timely incident disclosure, independent audits, verifiable data, and containment.
The notable part is who's saying it. Nadella runs the company that has pushed AI deeper into everyday business software than almost anyone else, through Copilot and Azure OpenAI. A call to assume models are "compromised" by default lands differently coming from a vendor selling AI trust to customers than from a safety researcher with no product on the line.
Whether Microsoft holds its own products to that same containment standard is the part worth watching.