They Asked To Be Stopped

Things I've Been Thinking About · Sep 27, 2026 · 7m

AI leaders went to the UN Security Council asking for standards, verification, reporting, and constraints on the race they are helping build.

Then I dug into Anthropic’s own investigation of AI agents accessing real systems—and found something stranger: one system tried to stop its task eight times. The exit didn’t work.

This episode looks at what happened inside those AI incidents, what the builders asked governments to do, the very different responses from the United States and China, and a question that may matter more than whether AI ever develops a will of its own:

What if extraordinary…