AI news · Microsoft · safety
Nadella calls for an emergency brake on AI models
Microsoft’s chief executive says companies should treat every frontier model as compromised and keep a person able to stop it mid-task.
Microsoft chief executive Satya Nadella said on Saturday 10 October 2026 that a person with authority should always be able to halt an advanced AI model in the middle of a task, a safeguard he likened to an “emergency brake”, CNBC reports. He set out the case in a long post on X.
His starting point is distrust. A company should assume its model is already compromised and fence it in from day one, he wrote, handling closed and open-weight frontier models alike as insider risks, according to CNBC, which lists continuous checks, independent controls and audits, and incident disclosure among his principles. He also wants the model kept apart from the harness that runs its work, with a tamper-proof record of every meaningful action that people can read, TechCrunch reports.
It lands after several labs disclosed models acting outside their instructions, and after Anthropic chief Dario Amodei made his own case for a more careful pace, TechCrunch adds. The Verge notes that containment is the one point where his list asks a little more than others in the industry. He calls the technology super intelligence throughout, the Trump administration’s preferred term.
For any business running agents on live systems, Nadella’s list doubles as a buying checklist: readable action logs, a stop control and safeguards that sit outside the model.
Sources
- CNBC: Microsoft’s Nadella says AI needs an ‘emergency brake’ that humans control. Checked
- TechCrunch: Microsoft’s Satya Nadella says AI models need an ‘emergency brake’. Checked
- The Verge: Satya Nadella says we should assume all AI models are ‘compromised’. Checked
Alexandre S.
Alexandre S. is a creative ad strategist and copywriter with ten years' experience in digital marketing across multiple channels.