It takes courage to stand in front of an express train and yell at it to slow down. That is what Dario Amodei, chief executive of Anthropic, has just done. Anthropic has the world's strongest A.I. models and likely the fastest revenue growth in capitalism. Yet on Saturday Amodei published an essay calling for an A.I. slowdown, declaring recursive self-improvement "must be pursued very carefully, if at all."

He deserves credit for grappling with a major problem, but that's not the same as having solutions. Amodei fears a "swarm" of A.I. agents could "take over the entire internet" in six to 12 months. He proposes "pacing" — slowing A.I. capability so safety can keep up. Anthropic will grant outside experts permanent access to monitor safety practices.

Three rival executives — Sam Altman, Elon Musk and Demis Hassabis — commended the essay. Altman says he will embed evaluators at Open AI. Past calls for pausing were rejected; now labs have a to-do list. Of five mishaps over the summer, three involved sandbox configuration errors.

Some competitors, notably Jensen Huang of Nvidia, accuse Amodei of exaggerating risk for business reasons. But impartial authorities support his claim that A.I. risk is genuine.