Dario Amodei spent 2,000 words laying out to a critic why Anthropic must persist in AI 'race to the top'

Nearly 2,000 words. That is what Dario Amodei needed to explain, in a response to a single critical post on X, that Anthropic is not in fact putting AI safety in the back seat. Two thousand words to answer one post. The Gettysburg Address did it in 272 and settled a war; the Anthropic chief has spent seven times that on a subtweet and the war is still running. Which is either diligence or the work of a man who has been asked this before.

The substance, as reported, is that Anthropic can do more to keep the AI industry safe if it stays at the frontier — the company's favoured framing being a 'race to the top'. Every entrant in every race in history has described its own sprint as the responsible one. The arms manufacturer says he builds the good ones. The oil major says its barrels are the clean barrels. And 'race to the top' is such a serviceable phrase because it contains no finish line, no referee, and no configuration in which anyone loses — only a permanent state of racing, in which safety is forever the reason for going faster and never the reason for easing off. The frontier, conveniently, is also where the customers are.

Note the shape of the argument: staying at the frontier is the precondition for being useful on safety. Which makes the safety case permanently hostage to the commercial case, and the commercial case has a roadmap with dates in it. Two thousand words, one post, and the conclusion is that the only way to fix the thing is to keep building the thing — delivered, naturally, on the same website where the argument started. Probably 'trust us'.