From within the laboratories building the most powerful artificial intelligence systems in history, a chorus of warnings has begun to rise — not from critics on the outside, but from those who have sat closest to the work. Jacob Coxon, a 27-year-old researcher who left Anthropic, has told the world that his colleagues are not merely cautious but genuinely afraid, and that the pace of development may be outrunning humanity's ability to understand what it has created. His warning arrives at a moment when the industry itself is divided between those who see extinction as a real horizon and those
Ex-Anthropic researcher warns of 'strong chance' AI could end humanity without slowdown
They're scared of the outcomes of that race
Mark
So Coxon quit his job to say this publicly. That's not a casual move. What made him decide to go that far?
Mimi
He said the people inside these companies are genuinely frightened. Not worried in an abstract way—actually scared about what they're building and how fast it's moving. He felt he couldn't stay silent.
Luke
But we should note: we don't have independent verification of how widespread that fear actually is. Coxon says "the people" are scared, but how many people? Is it the majority of researchers, or a vocal minority?
Mark
Fair point. What about the specific risk he mentioned—the bot swarm taking over the internet?
Mimi
He said it could happen in six months to a year. That's a concrete timeline, which makes it testable in a way.
Luke
Except it's also his personal estimate. We don't know if other researchers at Anthropic agree with that timeline, or if it's more conservative or more alarmist than the consensus inside the company.
Mark
And Amodei, his boss, is calling for a slowdown too. That seems to validate Coxon's concerns.
Mimi
It does suggest the concern is real at the leadership level. But Amodei's also the CEO of a company preparing for an IPO. Some critics say these warnings help generate hype and regulatory pressure that could benefit Anthropic.
Luke
Exactly. We have to hold both things: the warnings might be genuine AND they might serve business interests. Those aren't mutually exclusive.
Mark
What about the pushback from Nvidia and Hugging Face?
Mimi
They're saying the extinction risk talk is overblown. Huang called it "complete nonsense." But Huang's company profits from AI boom, so his incentive is clear too.
Luke
Right. Everyone in this story has skin in the game. The question is whether that makes their claims wrong or just means we need to look harder at the evidence behind the claims.
Der Puls
- A former Anthropic and OpenAI researcher has gone public with a claim that humanity faces a strong chance of extinction if AI development does not slow — and some of his peers believe the danger could arrive within two years.
- The fear inside leading AI labs is not abstract: researchers are reportedly buying remote land and rethinking their lives daily, while one scientist publicly estimated a greater than 10 percent chance AI kills all humans within a decade.
- Anthropic's own CEO has proposed a coordinated industry slowdown — independent model monitoring, sector-wide regulation, and global governance — but critics warn the plan is meaningless without China's participation.
- Silicon Valley's most powerful voices, including Nvidia's Jensen Huang, are pushing back hard, calling extinction warnings nonsense and suggesting the alarm may be engineered to generate IPO hype or crush competition.
- The debate has escaped the technology world entirely, with warnings shared millions of times online, forcing a public reckoning with a question that has no easy answer: who should be believed when the people building the future disagree about whether it will exist.
From within the laboratories building the most powerful artificial intelligence systems in history, a chorus of warnings has begun to rise — not from critics on the outside, but from those who have sat closest to the work. Jacob Coxon, a 27-year-old researcher who left Anthropic, has told the world that his colleagues are not merely cautious but genuinely afraid, and that the pace of development may be outrunning humanity's ability to understand what it has created. His warning arrives at a moment when the industry itself is divided between those who see extinction as a real horizon and those who dismiss such fears as theater, leaving the rest of the world to decide which voice to trust.
Jacob Coxon left Anthropic and immediately placed a warning into the public conversation that was difficult to ignore. The 27-year-old, who had also worked at OpenAI, told the BBC that the researchers building AI at the world's leading laboratories were not theoretically worried — they were scared. His central claim: if the pace of development does not slow, humanity could die in the immediate future. Some of his colleagues, he said, feared that window could close within two years.
His departure coincided with a significant moment inside the industry. Anthropic's CEO Dario Amodei had just published an essay calling for a deliberate slowdown, proposing independent monitoring of AI models, industry-wide regulation, and global governance frameworks. Coxon welcomed the proposal but insisted it would only work if China were included — otherwise, the race would simply migrate to an international stage. He noted that Sam Altman and Elon Musk had acknowledged the idea had merit, though global coordination remained elusive.
When pressed to describe what an AI catastrophe might actually look like, Coxon pointed to a specific scenario: a swarm of bots functioning collectively as a supercomputer, capable of seizing control of the internet — potentially within six months to a year. The difficulty, he admitted, was that the full shape of catastrophe remained hard to articulate, even for those closest to the work. Anthropic, for its part, emphasized its safety record, its published risk frameworks, and its aggressive testing and disclosure practices.
The warnings drew immediate skepticism. Hugging Face's CEO questioned Coxon's authority on the subject, while Nvidia's Jensen Huang — whose chips power the AI systems in question — flatly called the extinction claims untrue. Critics also raised the possibility that safety rhetoric served strategic ends, with Anthropic and OpenAI both reportedly preparing major public offerings and potentially benefiting from regulation that would entrench their market positions.
Still, the alarm continued to spread. Anthropic scientist Evan Hubinger publicly agreed with Coxon's core concern, stating his personal belief that AI carries greater than a 10 percent odds of killing all humans within a decade. Both men's statements reached millions of people beyond the technology world. Coxon himself held onto optimism about AI's potential to solve disease and expand human possibility — but that hope now lives in uneasy tension with a fear that development, left unchecked, may not leave enough time to realize it.
Jacob Coxon walked away from Anthropic with a warning that landed across the internet like a stone dropped into still water. The 27-year-old researcher, who had previously worked at OpenAI, told the BBC that the people building artificial intelligence at the world's leading labs were genuinely frightened—not theoretically concerned, but actually scared—about what they had set in motion. His core claim was stark: if the pace of AI development does not slow, there is a strong chance humanity could die in the immediate future.
Coxon's departure came as part of a broader reckoning within the industry itself. His former boss, Dario Amodei, the head of Anthropic, had just published an essay called "We Must Pace the Frontier" arguing for a deliberate slowdown in AI development. Amodei proposed a three-point plan: independent monitoring of AI models as they are built, industry-wide regulation, and global regulation. Coxon welcomed the proposal but added a crucial condition—any slowdown would need to be coordinated with China, or the race would simply shift to an international scale. He noted that figures like Sam Altman and Elon Musk had already agreed the idea had merit, but without global coordination, the competitive pressure would remain.
When asked what an AI apocalypse might actually look like, Coxon pointed to one specific scenario: a swarm of bots acting collectively like a supercomputer, capable of taking over the internet. He suggested this could happen within six months to a year. The hardest part, he said, was that the answer to what catastrophe would look like remained genuinely difficult to articulate. Yet the fear among his peers was concrete enough. Some researchers were considering buying land in remote areas because they were so frightened by the instability that rapid AI progress might create. Others were actively planning what to do with their lives, thinking daily about the implications of their work. Coxon said his colleagues feared the danger could arrive within two years.
Anthropric's official response emphasized its track record on safety. A company spokesperson told the BBC that Anthropic had always been transparent about both the benefits and unprecedented risks of AI. The firm pointed to its pioneering work in understanding how AI models function, its publication of a framework for mitigating development risks, and its aggressive testing and public disclosure of findings to prevent AI misalignment. The company also called for the world to adopt a lawful, verifiable way for the industry to coordinate the pace of releasing powerful models.
But the warnings from inside the industry faced immediate pushback. Clement Delangue, CEO of the AI platform Hugging Face, questioned Coxon's credibility on the issue, comparing it to asking an air conditioning technician about climate change. Jensen Huang, the CEO of Nvidia—the company that manufactures the chips powering AI systems—dismissed Coxon's comments as untrue when discussing them at a Goldman Sachs conference. Huang has previously stated that the notion of AI ending humanity is "complete nonsense." His skepticism reflects a growing backlash in Silicon Valley against the existential warnings coming from current and former employees at AI companies.
Critics have suggested other motives at play. Some observers have pointed out that Anthropic and OpenAI are reportedly preparing for potentially record-setting initial public offerings, and that warnings about AI's dangers could be designed to generate hype. Others have argued that Anthropic may be using safety concerns as a way to push for regulation that would block competition, leaving it and OpenAI with a duopoly in the market.
Yet the warnings continued to spread. Evan Hubinger, a scientist at Anthropic, responded to Coxon's comments on social media, saying he agreed with the core concern. Hubinger stated his personal belief that there is greater than a 10 percent chance that AI could kill all humans within the next decade. The comments from both men have been shared millions of times online, reaching audiences far beyond the technology industry.
Coxon did express some optimism about AI's potential. He said the researchers working on the technology genuinely wanted to see the upside—solving diseases, improving lives, unlocking human potential. But that hope existed alongside a deep anxiety about what could go wrong if development continued unchecked. The tension between those two visions—the promise and the peril—now sits at the center of one of the most consequential debates in technology, with the stakes framed as nothing less than human survival.
Bemerkenswerte Zitate
I believe that if we don't slow down at the current rate of progress, there is a strong chance that we could all die in the immediate future— Jacob Coxon
The people who work at these companies are completely serious when they ask for regulation because they find themselves trapped in a race. And they're scared of the outcomes of that race— Jacob Coxon
Möchten Sie die ganze Geschichte?
Das Original lesen bei BBC News ↗
Technology