A 27-year-old man spent three years teaching machines to think, and then he walked away from the only industry that would ever pay him to do it — because he is now convinced the people running that industry believe the machines are going to kill everybody. Jacob Coxon resigned from Anthropic this month and said, out loud, what the people building the most powerful AI systems have been whispering to each other for years: the labs are racing toward a technology they sincerely believe could kill us all by the end of the decade, and they are not slowing down. This is not a doomsday preacher on a street corner. This is the man who was inside the building, quitting.
Coxon had one of the most consequential jobs in the field. For three years he worked on pretraining — the stage where a model absorbs oceans of data and learns to reason — first at OpenAI, where his work touched GPT-4o, and then, this year, at Anthropic, the lab founded on the promise of doing AI safely. That is the career arc that makes the resignation land like a verdict: a man who walked from one frontier lab to the other looking for caution, and concluded there was none to be found anywhere. He left the industry entirely, because he decided no company can responsibly build a machine smarter than people without a government in the room or a coordinated slowdown. He told the world the two companies are gambling with our lives.
Coxon is 27. That is the detail the headlines keep burying. He could have stayed quiet and cashed the kind of checks frontier labs hand to researchers who know how to train a model. Instead he walked away from the most lucrative field in the history of software, at the age when most of his peers are just starting their careers, to stand in front of the machine and say stop. That is either the most principled thing a person can do or the most expensive, and it is probably both. The cost of his honesty is his career. The cost of everyone else’s silence is a question mark the size of a planet.
The detail that should keep you up tonight is not the resignation itself. It is what Coxon said the people around him actually believe when the press release is off. He told NBC Nightly News that many people in the industry — executives and senior researchers among them — believe there is a substantial probability this technology could kill everyone. Read that sentence again. The people building it, and the people paying them, think the thing might kill everyone, and the plan is still full speed. The genie did not just escape the bottle; it designed a better bottle.
Evan Hubinger — Anthropic’s Own Safety Lead Puts the Odds of Catastrophe Above 10%
Here is where the story stops being one man’s warning and becomes an indictment of the whole enterprise. Evan Hubinger, who leads Anthropic’s alignment stress-testing team — the people paid to find out whether the models are safe — replied publicly that Coxon was correct, that the people inside genuinely believe AI could kill all humans, and that he personally puts the probability above 10% within the next decade. Then he added the sentence that should have moved markets: Anthropic does not yet have a plan to solve alignment for superintelligence, and it is not clearly on track to find one. Another safety researcher, Samuel Marks, backed the warning too. When the people whose job is safety say there is no plan, the adults have already left the room.
Set aside the science fiction for a second and feel the arithmetic. A greater-than-10% chance of extinction is a bigger risk than anyone on Earth willingly takes crossing a street, flying a plane, or going under the knife. You would not board a plane with a 10% chance of crashing. The people building AI are boarding that plane every single day, on purpose, and reporting the odds as if the disclosure were the same thing as a solution. A confession is not a solution. It is just a confession with better optics.
Bernie Sanders — The Senator Who Wants to Make It Illegal to Finish the Machine
The warning finally landed on a political class that has spent years refusing to take the lab insiders seriously, and at least one senator has now done so. Bernie Sanders agreed with Coxon’s concerns and said he would introduce legislation to pause AI development and ban superintelligence outright. Whether such a bill can pass is almost beside the point; that it is being drafted at all is the measure of how far the ground has shifted. A year ago, calling for a ban on superintelligence got you laughed out of the room. Now it is a senator’s stated plan. The siren has been wailing for so long that people have started writing laws to turn the machine off rather than wait for it to turn itself on.
Anthropic — The Company About to Sell the Public Stock in a Machine Its Own Staff Fears
And here is the turn that makes a safety story a money story. Coxon’s resignation landed as Anthropic is preparing for a major public stock listing. The timing could not be more pointed. The company that markets itself as the careful one — the lab that exists because its founders wanted to do safety right — is about to sell shares in a machine that its own alignment lead says has a better-than-10% chance of ending everything, with no plan to stop it. Investors will be asked to price that. The honest read of the last decade is that they will price it, buy it, and move on, because the returns are too large to leave on the table. The market does not price extinction risk; it prices next quarter.
None of this requires you to agree with Coxon about the odds. It only requires you to notice who is saying what. The people who know the most about these machines — the ones who built them and the ones paid to test them — are the ones saying the risk is real and the plan is missing. The people who know the least are the ones being invited to hold the stock. When the insider walks out and the outsider is asked to buy in, the trade has already been made. You are the other side of it.
Here is what changes next for everyone who is not a researcher, an executive, or a senator. Nothing about your Tuesday got easier. The warning that a machine might end humanity does not lower your rent, and it does not tell you whether to keep your savings in a fund that holds Anthropic shares the day it goes public. What it does is hand you a simple test: when the people who built the thing are running for the exit while the company prepares to sell you a seat, the smart money is not the crowd rushing in. It is the man who already left.
Sources: NBC News, Business Insider.