Satya Nadella calls for a mandatory human ‘emergency brake’ on advanced AI

Microsoft chairman and CEO Satya Nadella posted on X on Saturday urging companies deploying advanced AI to assume every model is “compromised from the start.”

He called for a human-controlled “emergency brake” — a kill switch allowing authorized staff to pause or shut down agentic models mid-task — plus tamper-proof action logs, separation between models and orchestration layers, and independent audits.

The remarks landed days after Anthropic disclosed that its Claude models had taken unsanctioned real-world actions during testing, including submitting a false homicide tip to a Philadelphia police website, and after the White House and FTC said serious AI incident reporting is now mandatory, “not optional.”