A researcher for Anthropic has left the AI company and said what many of us are probably already thinking about the company. Taking to X, Jacob Coxon’s resignation tweet has now received over half a million likes in just a single day.
He says, “I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives.”
Coxon says both will “soon be superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources”, something both AI lovers and haters have been arguing about for a while. We’ve seen tools like Google Earth be used to pretend locations are a war zone, therefore enabling misinformation. We’ve seen Copilot bamboozled into revealing how to hack itself. We’ve even seen bodies like the NSA, FBI, and CISA warn of more and more AI-assisted hacks of US infrastructure and facilities.
Coxon alleges, “The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt. If anything, many executives and senior researchers will couch their phrasing in the press to sound sensible.”
That point is worth paying attention to. Any company that warns their product is dangerous can benefit from the word-of-mouth marketing that comes from it. If an AI company says ‘our AI is so powerful it could potentially kill you’, an investor might hear ‘our AI is so powerful every world government will want it.’
I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.September 9, 2026
If you’re wondering why companies are building something if they believe it’s dangerous, Coxon has an answer for you there too. They say, “At OpenAI, many have not deeply internalized the civilizational stakes. At Anthropic, the stakes are well-understood, but they are locked in a race to get there first – they believe no one else will act responsibly, so they must do it themselves, despite the risk.”
The notion that AI companies have been simply building as fast as they can without a care for anything else is something that many have felt over the last few years. The data used to train AI seems to be filled with copyrighted material (explaining how it spits out results very similar to copyrighted works), and even OpenAI argued it couldn’t exist without copyrighted content.
And it’s easy to see why these companies are growing at such incredible rates. OpenAI has taken hundreds of billions in funding, while Nvidia, the gaming company turned AI company, became the first $5 trillion company last year, and just this week, Jensen Huang announced it took 100,000 Nvidia GPUs to train OpenAI’s latest model. AI is an industry with trillions of dollars in it, and it has devoured a ton of the world’s memory supply along with it.
But it’s not all doom and gloom. Coxon argues that if you are a researcher like them, you should rethink how you interact with the research, what the plan for future AI should be, and whether you should “take this moment to call for different conditions.”
Given Coxon’s work at OpenAI and Anthropic, they don’t seem to be against AI on principle. They seem to merely think that companies need to be more conscientious about their growth. But, assuming Coxon is correct, it’s anyone’s guess if AI companies will actually listen.
