|
Getting your Trinity Audio player ready...
|
Jacob Coxon spent three years working on pretraining research at two of the most prominent artificial intelligence companies in the world, OpenAI and Anthropic. On September 8, he resigned from Anthropic and published a series of posts on X that quickly became one of the most widely viewed statements ever made by a departing AI researcher, amassing more than 90 million views within 24 hours. His message was blunt. Neither company, he wrote, is acting responsibly, and both are racing straight toward self-improving superintelligence while gambling with our lives.
Coxon, 27 and British, told TIME that no single incident or discovery triggered his decision to leave. Instead, he said his resignation came down to two conclusions that had become impossible for him to ignore: that the pace of AI development is clearly speeding up, and that it is not under control. In his public posts, he warned that upcoming AI systems would soon become superhuman, capable of hacking almost anything, transforming entire fields overnight, and accumulating real power and resources, adding that progress in each of these areas shows no sign of slowing.
What made Coxon’s departure notable within the industry was not just the warning itself but who was making it. Many of the most prominent researchers who have left frontier AI labs with public safety concerns previously worked specifically on alignment or safety teams. Coxon, by contrast, spent his career building the very capabilities he now says frighten him, having conducted pretraining research, the work that goes into building a model’s raw capabilities, at both companies rather than working on the safety side of the business.
Perhaps the most striking claim in Coxon’s public statement concerned what he says people inside these companies privately believe. He wrote that the people building AI earnestly believe it could kill everyone by the end of the decade, and argued this was not a marketing stunt, noting that executives and senior researchers often soften their public language even as they express the same fears privately. In a separate interview with the Wall Street Journal, Coxon said the industry is on track for some of the most aggressive risk scenarios, in which things could spiral out of control as soon as next year, and pointed to competitive pressure, both between OpenAI and Anthropic and against Chinese AI developers, as a force that makes safety trade-offs feel unavoidable to the people making these decisions day to day.
Coxon’s claims did not go unanswered inside Anthropic. Evan Hubinger, a current Anthropic employee working on alignment science, responded publicly on X that he personally estimates the probability of a catastrophic AI outcome at greater than 10 percent within the next decade, effectively confirming that some current staff share the concern Coxon attributed to the field more broadly. According to reporting from TIME, Coxon said he discussed his decision with many colleagues before leaving and found broad, if largely private, agreement that the industry’s current trajectory carries serious risk. He described what he called an atmosphere of resignation among some researchers, where people accept that the competitive race is happening regardless of what any one person does, and conclude that the most useful thing they can do is try to make their own individual work as safe as possible, even while suspecting the broader effort could still spiral out of control.
Coxon’s exit was not an isolated event inside Anthropic that week. Mrinank Sharma, a member of Anthropic’s technical staff since 2023, also announced his resignation around the same time in an open letter to colleagues, writing that the world is in peril from a series of interconnected crises, not limited to AI alone. Separately, Jakub Pachocki, OpenAI’s chief scientist, had cautioned just a day before Coxon’s post that the world remains unprepared for how quickly AI capabilities are advancing and said he expected developers to eventually slow their own work voluntarily. Former OpenAI researcher Daniel Kokotajlo, appearing on The Joe Rogan Experience, offered a similarly stark framing, saying that if the current race continues, humanity risks losing control of AI systems entirely.
Anthropic has publicly maintained that it takes these risks seriously precisely because it continues developing frontier AI, arguing, according to Coxon’s own account of internal conversations, that the company believes competitors will not act responsibly on their own, so it must remain at the frontier to influence how the technology develops rather than ceding that ground entirely. That framing, that the safest path runs through staying competitive rather than stepping back, has become one of the central points of tension in the broader debate Coxon’s resignation reignited, since it implies that even researchers who privately fear catastrophic outcomes see few realistic alternatives to continuing the race.
Not everyone in a position of influence has treated the warnings as a reason to slow down. President Donald Trump dismissed calls for an industry-wide pause on AI development, arguing that any slowdown would risk ceding America’s competitive advantage over China in the global AI race, a view that highlights how thoroughly geopolitical competition has become intertwined with questions of AI safety at the highest levels of government.
Coxon’s public statement, and the swift, partial confirmation it drew from a current Anthropic employee, has added fresh urgency to a debate that has increasingly moved from academic conferences and research papers into public view over the past year. Whether his warning translates into any meaningful shift in how frontier AI companies operate, or simply becomes another widely shared but ultimately unactioned moment in an already crowded field of AI safety warnings, remains an open question that will likely depend on decisions being made inside these companies over the coming months rather than any single resignation post. Continuing coverage of how AI safety debates are shaping the technology industry is available on Business Tech. Coxon’s full interview and further detail on his reasoning are available through TIME’s original reporting, and additional background on Anthropic’s stated approach to AI safety can be found on the company’s official site.
