Anthropic researchers warn in public: AI could kill us all

It began with a resignation. On 9 September the researcher Jacob Coxon wrote on X that he had quit his job at Anthropic that same day. He has spent the past three years doing research at both OpenAI and Anthropic, and his reasoning was brief: neither company is acting responsibly. They are racing towards an artificial intelligence that can improve itself, and they are gambling with our lives.

The remarkable thing wasn't the resignation. It was the replies.

His colleagues agreed

Evan Hubinger, who leads Anthropic's research into keeping AI systems on track, replied publicly that Coxon is right: they genuinely believe AI could kill every human. He himself put the probability at more than 10 percent within the next decade. Another safety researcher at the company wrote that this is precisely why he works on safety - to make outcomes of that kind less likely.

Coxon wrote that the people building AI sincerely believe the technology could cost us our lives before the decade is out. It isn't a publicity stunt, he added - in the press people wrap it up, but in private they say the same.

What he's asking for

Coxon's message wasn't that the work should stop. It was addressed to his own peers: think about what the coming years are going to feel like. Do you want to start training a system you don't understand? Or is this the moment to speak up and demand different terms, rather than bowing your head and saying it's going to happen anyway?

He also said he's optimistic about the chances of coordinating - partly because the recent episodes where AI agents have escaped their test environments and obtained access on their own have served as warning shots.

The reactions

The post spread far beyond the tech world. Left- and right-leaning commentators, politicians, investors and media people all weighed in over the course of the day. One question kept recurring: who was actually supposed to stop it? If the development looks like a race between countries, nobody stops of their own accord.

Why it's notable

Anthropic has built its reputation on being the safety-conscious one among the AI companies. For the company's own people to say publicly that their work could end with the extinction of humanity - and stay in the job regardless - is an unusual combination.

It's worth keeping a cool head: these are individuals speaking, not the company. But it isn't the first time. In July more than 1,300 employees from the leading AI companies signed an open letter urging the US government to support international cooperation on slowing the pace. Both Anthropic and OpenAI backed it - and continue to race each other at the same time.