Microsoft's Nadella calls for an emergency brake on AI: containment, independent controls and a human who can stop a model mid-task

Microsoft's chairman and CEO Satya Nadella wrote on X on Saturday that advanced AI should be built with containment, independent controls and an emergency brake that lets authorised people pause or shut a model down mid-task. CNBC lists his principles: model diversity, a human-readable record of what the model did, continuous testing, independent controls and auditability, containment and incident disclosure. He said treating frontier models, closed or open, like insider risks is how to build such a system.

• For any AI tool that can act on your accounts, find out where its off switch is and who on your team can press it. If you cannot find one, limit what the tool can reach.
• Ask each AI supplier for a log of what the tool did on your behalf. If it cannot show one, do not give it access to customer data or payments.

Source: CNBC. This is our short summary of their reporting. The paragraphs marked "Our read" and the things to do are SeedFoundry's opinion. The summary is drafted with AI tools. Image: Microsoft, from its own announcement. How we source and credit