Anthropic Researcher Warns There’s 10% Chance AI Could Kill All Humans’ By Next Decade
A growing debate over the future of artificial intelligence has taken a darker turn, with a senior Anthropic researcher warning that some people inside the company genuinely believe advanced AI could one day wipe out humanity.
The warning comes after another researcher, Coxon, reportedly left Anthropic over concerns that the company is moving too aggressively toward building self-improving AI systems without fully addressing the risks that could come with them.
According to The Wall Street Journal, Coxon’s departure was driven by fears that the race to develop increasingly powerful AI could push the technology into dangerous territory. He told the newspaper that the world could be heading toward “a lot of the most aggressive” scenarios, warning that by the end of next year, things could potentially already be “out of control.”
Coxon also took his concerns to X, where he compared his experience at Anthropic with his time at OpenAI, the company behind ChatGPT.
According to Coxon, Anthropic employees have a much stronger understanding of what he described as the “civilizational stakes” surrounding advanced AI. But that awareness, he argued, has not necessarily translated into slowing down.
Instead, Coxon said Anthropic appears to be caught in a race to develop the technology first because of a belief that other companies may not act responsibly. In his view, the logic is essentially that Anthropic has to move forward because it cannot trust competitors to do the right thing, even if doing so comes with enormous risks.
That tension is at the heart of a much larger conversation happening across the AI industry: how fast is too fast?
The concerns come as several leading AI researchers and executives have increasingly called for more caution in the development of frontier AI systems.
In July, a group of prominent figures signed a statement titled “Pacing the Frontier,” arguing that the industry, governments and society may need the ability to slow down long enough to address emerging dangers, develop stronger safety measures and improve oversight.
Among those who signed were Anthropic co-founders Dario Amodei and Jared Kaplan, OpenAI Chief Scientist Jakub Pachocki, Meta AI chief scientist Shengjia Zhao and other major figures in the field.
The statement acknowledged the enormous potential of AI but also pointed to a difficult problem: even companies that want to slow down may feel they cannot afford to do so if their competitors continue racing ahead.
That concern was echoed again by Pachocki in a blog post on Sunday. He argued that no AI laboratory has yet solved alignment and monitoring well enough to justify continuing to scale increasingly powerful systems at maximum speed for much longer.
In other words, the people building the technology are increasingly warning that the AI race could become a case of everyone hitting the accelerator while simultaneously worrying about where the road leads.
And with researchers now leaving companies over those concerns, the debate is no longer simply about what AI might become in the distant future. For some insiders, the question is whether the industry is already moving faster than its ability to control what comes next.
