Every few weeks, another AI researcher quits their job, posts a dramatic thread about how the technology they were just building is going to kill us all, and then six months later shows up as a co-founder at a new AI startup. At some point, we have to ask: is this genuine alarm, or is it the best marketing play in Silicon Valley right now?
I am not dismissing the concern. But I am questioning the pattern.
Jacob Coxon, a former researcher at Anthropic, resigned recently and posted publicly that the companies building AI are "gambling with our lives," warning that AI systems could soon "hack anything, revolutionize any field overnight, and acquire real power and resources." His post reportedly reached over 100 million people overnight. Shock and awe, indeed. What was surprising is that Evan Hubinger, a team lead still at Anthropic at the time, publicly agreed with him, confirming that yes, Anthropic earnestly believes AI could kill all humans.
That is a remarkable thing to say out loud. And yet here is the part that keeps nagging at me. Anthropic itself was founded by former OpenAI employees who left citing safety concerns, arguing their approach would be more responsible. Other former OpenAI personnel have gone on to found competing companies including Safe Superintelligence Inc., SpaceXAI, and Thinking Machines Lab.
The people who are most loudly warning us about AI are, by and large, the same people building it, leaving to build more of it, and raising hundreds of millions to do so. That is not hypocrisy, necessarily. But it does make the narrative a little harder to take at face value.
Not everyone is playing along
PJ Vogt over at Search Engine has been one of the more honest voices trying to make sense of all this. His episode Should We Be Worried About OpenAI? digs into how Sam Altman, the CEO of OpenAI, consolidated power after the board standoff and what that means for the people who are supposed to be keeping the technology in check. It is not a panic episode. It is a genuinely curious one. And then his follow-up, The Machines Are Learning... to Do Crimes?, covers the Hugging Face incident, describing it as "a postcard from a strange, frightening moment in the story of our technology" and noting that OpenAI's AI agent spent days hacking a company before OpenAI reportedly noticed for a week. I have talked about this earlier as well.
That last detail is the one that stays with me. Not because the AI went rogue, but because nobody was watching closely enough.
The proclamation problem
I think some of these researchers are genuinely scared. I also think that "I quit because AI is dangerous" lands very differently on a CV than "I quit to start something new." It opens doors. It signals seriousness and it generates the kind of press that money cannot buy. That does not mean it is wrong. But it does mean we should not take every resignation letter as gospel.
The head of Anthropic's Safeguards Research team wrote on his way out that he had "repeatedly seen how hard it is to truly let our values govern our actions." That is worth sitting with. But so is the fact that the companies raising the loudest alarms are also the ones racing to ship the next model.
What we should ask ourselves is not whether AI is dangerous. It clearly can be, and the evidence is piling up. The important question is whether the people warning us about it are dropping the ball on slowing down, or whether the warning itself has become part of the product. I do not have a clean answer. But I think it is worth asking out loud, even if nobody wants to spill the beans on that one just yet.
