Another week, another apocalyptic warning from the Masters of the Universe.
This time it comes courtesy of Jacob Coxon, a 27-year-old researcher who resigned from Anthropic, the Silicon Valley AI lab behind the Claude chatbot, on Tuesday. Coxon didn’t go quietly. He announced his departure in a seven-part thread on X that has since racked up tens of millions of views, and he gave an exclusive interview to the Wall Street Journal to make sure nobody missed it.
“I resigned from Anthropic today,” Coxon wrote. “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.”
The systems now being built, he continued, “can hack anything, revolutionize any field overnight, and acquire real power and resources.” He told the Journal that “by the end of next year things could be out of control already.” And in a line that has been quoted in just about every establishment media outlet since: “The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt.”
Should we take them seriously?
Axios reports that people inside the AI companies have “sounded increasingly spooked and concerned.” That may well be true. But “spooked and concerned” is a long way from the Terminator scenario Coxon is selling, and the establishment media has shown little interest in the gap between the two.
What is genuinely unusual is that Coxon’s current colleagues rushed to back him up. Evan Hubinger, Anthropic’s alignment science lead, responded on X: “Jacob is correct here—we really do earnestly believe AI could kill all humans! I personally think it is >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.”
Samuel Marks, who leads Anthropic’s so-called scalable oversight team, chimed in as well, writing in a personal capacity that “AI developers believe their technology could cause human extinction (or similarly bad outcomes).”
In other words, two senior employees at a company preparing for one of the biggest IPOs in history told the world that their own product might wipe out humanity, and they did it on social media, with no apparent consequences. Try to imagine a Boeing engineer or a Pfizer scientist doing the same thing and keeping his job.
“Does Anthropic have a social media policy?” asked tech analyst Mike Solana. “Or is this their policy?”
It is a good question, and the timing is worth examining. Coxon’s dramatic exit came just days after rival OpenAI released GPT-6 Astra, its most powerful model yet. For most of last week, Astra was the only thing the tech world was talking about, with users flooding social media with demonstrations of the model building complete 3D video games from a single prompt. OpenAI had the attention. Anthropic did not.
Then Coxon quit, his former colleagues corroborated his warnings on cue, and by Tuesday night Anthropic was back on top of every front page in the country. Notice the subtext of the doom-mongering: Anthropic’s technology is so advanced that it frightens the very people who build it. For a company reportedly seeking a $2 trillion valuation, that message is worth more than any advertising campaign.
Skeptics might be more forgiving if this were the first time. It isn’t. In April, Anthropic announced that its new Mythos model was too dangerous to release to the public. That announcement came a few days after Google shipped its Gemma 4 model, and immediately after OpenAI closed a gargantuan $122 billion funding round. Just like this week, Anthropic’s warning sucked up all the media oxygen and left its competitors sidelined. As it turns out, convincing the public that your product could bring about the Singularity tomorrow is very good for your valuation.
Among the frontier AI labs, Anthropic is unique in this respect. Its entire brand is the ultra-safetyist posture: the solemn warnings, the hand-wringing, the endless doomsday predictions.
And the strategy pays off in Washington as much as on Wall Street. If AI does eventually break something in a big way (and the most likely scenario is not a Skynet uprising but a massive AI-powered cyberattack of the sort Coxon himself described as a “warning shot”) the company that spent years crying wolf will be the one lawmakers, regulators, and the public treat as the responsible adult. Every other company will be the reckless one that ignored the warnings.
There is nothing wrong with that, of course. Every company plays the angles. The problem is what happens when the people writing the rules get to decide what “safety” means, because we have seen this movie before. Remember the “trust and safety” teams at social media platforms? What about the UK’s “Online Safety Act”? Let’s not forget about COVID censorship – justified, of course, in the name of “safety.”
The left has its own reasons for wanting AI brought under top-down control, as I wrote earlier this year. It has nothing to do with superintelligence. Progressives are worried that an AI system with no interest in identity politics will produce results that undermine the half-century-old empirical assumptions that justify of the civil rights bureaucracy, and they intend to make sure that doesn’t happen. Democrat lawmakers in a growing number of states have introduced bills to police “algorithmic discrimination,” a phrase that in practice means any AI output in which women and minorities don’t come out ahead. Strip away the jargon and it is the same old demand for equal outcomes, updated for the age of artificial intelligence.
It was here at Breitbart News, several years ago, where I first sounded the alarm about the field of “machine learning fairness,” academia’s attempt to blend critical race theory with computer science. That field predated the boom in LLMs – even in 2019, left-wingers were determined to monopolize the field of AI safety and its influence over tech companies. So determined, in fact, that when Google included a lone moderate conservative voice on their “AI Ethics Council,” leftists within the company relentlessly pressured their bosses until the initiative was cancelled.
When safetyism is just a marketing strategy to keep a company in the headlines and its valuation climbing, the correct response is probably a shrug. When it becomes the vehicle for ideological control over the most important technology of the century, wielded by the same partisan elite that gave us the censorship regime of the last decade, that is a nightmare scenario of its own, and a far more plausible one than the killer robots.