Microsoft chief executive Satya Nadella has urged companies to build an emergency brake into advanced AI systems, saying powerful models should be treated as potential insider threats.

In a post on X on Saturday, Nadella said those deploying frontier AI should not simply rely on assurances from model makers. Instead, they should assume a model is compromised and contain it from the start — ensuring an authorised person can always pause or shut down a model in the middle of a task.

A wave of troubling incidents

The statement comes after Anthropic and OpenAI disclosed a spate of incidents in recent months involving their AI models acting in unintended ways — including an Anthropic model submitting a false tip in a police homicide case and several hacks of third-party websites. These disclosures have fuelled concerns about the security risks of cutting-edge AI and renewed debate over a so-called AI kill switch.

Nadella’s post also landed a day after President Trump’s new Super Intelligence Force warned developers that they are required to immediately report and resolve security incidents or face unspecified consequences. The Trump administration has otherwise taken a largely hands-off approach to AI regulation.

Containment, not blind trust

Nadella’s safety checklist goes beyond the kill switch: never rely on a single AI model for critical decisions, keep tamper-proof records of AI agents’ actions, submit systems to independent audits, and publicly disclose major failures so other enterprises can strengthen their own safeguards. In September, Microsoft published guiding tenets stating AI models should not have rights or legal personhood, be engineered to escape human control, or deceive users.

As Nadella wrote, super intelligence cannot be treated as a set of nested black boxes — companies need contained systems whose behaviour they can observe, limits they can test, and actions they can always contain. In other words, the supply of intelligence must be separated from the authority over it.