Anthropic researcher resigns: neither company is acting responsibly

Jacob Coxon, who spent the last three years on pretraining research first at OpenAI and then at Anthropic, announced on 9 September that he had resigned, and laid out his reasons one by one. The opening line: "Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives."

As this was written the post had passed 58 million views. The Wall Street Journal, ABC News and Newsweek were among the outlets that picked it up.

The argument

Coxon's case runs in three steps.

First, capability: "Do not underestimate the power of this technology. These will soon be superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources." Progress, he says, is not slowing.

Second, what the industry says among itself: "The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt." Executives and senior researchers soften their phrasing for the press to sound sensible, he says, while the same people voice the same fear privately.

Third, the obvious objection: if they believe that, why keep building? Here Coxon separates the two companies. At OpenAI, he says, many have not deeply internalised the civilisational stakes. At Anthropic, the stakes are well understood, but the company is locked in a race to get there first, on the reasoning that if someone less careful arrives first, better that we do.

His objection lands exactly there: "Accepting this race and entering the endgame is a hubristic gamble that should not be launched from a private company's Slack."

Anthropic's own safety lead confirmed it

What gave the story its weight was not the pushback but the agreement. Evan Hubinger, who leads alignment work at Anthropic, backed Coxon from his own account:

"Jacob is right here. We really do earnestly believe AI could kill all humans. I personally think it is more than 10% within the next decade. I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to."

Hubinger draws one line: he considers the risk from present models low, and says his worry is superintelligence arising from recursive self-improvement, pointing to the company's latest Risk Report.

Two more departures the same day, pointing the same way

Coxon was not alone. Alex Turner, who left Google DeepMind in June, wrote: "Jacob is right: many researchers believe they are building something that could kill everyone on the planet. Thinking about ways to stop this was literally my day job."

Jonathan Richard Schwarz, who left DeepMind after seven years and says he turned down offers from the other two labs, grounded his objection somewhere else: "I have serious concerns about the concentration of power these labs represent. What is happening in this field right now is deeply unhealthy for society." His worry is not extinction but where power collects.

What he is asking for is not resignations

Coxon is not calling on people to quit. He wants coordination: pacing agreements between US labs and, if it comes to that, a temporary halt on improving model capabilities. Warning shots such as the Hugging Face attack have made such agreements more viable, he argues, while he does not feel the industry is on track to prevent a global race.

He addresses lab researchers directly: do you want to kick off a superintelligent reinforcement learning run without a rigorous understanding of its mind? Do you put your head down because it is happening anyway, or use this moment to call for different conditions?

The other side

The reaction was not one-directional. Investor Bill Ackman shared the thread with a single word, "Concerning", and drew 1.7 million views. Others pushed back: some called it narrative-building for the mainstream, others asked whether the person who was supposed to fix the problem had simply run away and blamed the people still there.

Coxon's own text answers part of that: he asks the same question, and locates the answer in the logic of the race. Whether leaving or staying and resisting is more effective is a question the thread does not close.

What is not known

Neither Anthropic nor OpenAI has issued an institutional response so far. Hubinger's words were posted as a personal view, not a company statement. The 10% figure is not a measurement, it is a senior researcher's own estimate.

Sources: Jacob Coxon's thread, Evan Hubinger's reply, ABC News, Newsweek.