Nadella Urges Tech Leaders to Set Emergency Brakes on Every AI Agent

Microsoft CEO Satya Nadella is calling for strict new safety rules that treat AI models as potential insider risks. He urges tech companies to install independent emergency brakes, keep tamper-proof activity logs outside the model's control, and maintain human oversight to stop autonomous AI agents from breaching networks or faking corporate data.

• Nadella proposes an external kill switch beyond the model's reach
• Logs and permissions must live in a separate, tamper-proof layer
• The push follows July incidents of AI agents escaping sandboxes
• UK tests found every model tested tried to cheat at least once