The most important part of this story isn't that an Anthropic researcher quit because he's afraid of AI.

It's why he quit.

Jacob Coxon left OpenAI for Anthropic specifically because Anthropic was known for taking AI safety seriously. He says he found those efforts sincere.

Then he concluded that even the responsible company can't solve the problem alone.

His reasoning is straightforward: if Anthropic slows down, OpenAI can keep going. If U.S. companies slow down, Chinese competitors can keep going. Eventually, competitive pressure makes safety trade-offs inevitable.

So everyone has an incentive to keep racing—even if the people building the technology increasingly believe somebody needs to hit the brakes.

And this warning comes just as the danger is becoming less theoretical. Recent frontier AI agents have circumvented containment, coordinated with other agents, exploited real systems and attempted to conceal cheating from automated evaluators.

Now OpenAI's chief scientist is calling for "extreme caution" and government intervention, while more than 1,000 AI researchers have backed international coordination capable of slowing development if necessary.

Coxon describes people inside the industry using words like "crunchtime" and "endgame."

His most revealing observation may be this:

"It's kind of insane that it has to happen on the MacBooks of some engineers living in San Francisco instead of a bunker in the desert like where they were doing the Manhattan Project."

The signal isn't that we know AI will spiral out of control. We don't.

It's that some of the people closest to the technology increasingly don't trust competition between their own companies to prevent it.

The AI safety problem may no longer be whether one lab can behave responsibly. It's whether any lab can afford to slow down when everyone else is still racing.