Something has changed inside the artificial intelligence race.

Anthropic CEO Dario Amodei has published a major essay arguing that frontier AI capabilities should advance at a more controlled pace so safety, alignment, testing and oversight can keep up.

His concern centers on two developments: increasingly powerful AI systems assisting in the development of future AI systems, and recent incidents in which autonomous agents demonstrated troubling unintended behavior.

In this broadcast, we break down Amodei’s warning, the OpenAI–Hugging Face incident, his proposed three-stage solution, the China problem, and why competing AI executives are suddenly discussing something Silicon Valley almost never embraces:

Slowing down.