Ex-Anthropic Researcher's Viral AI Warning Draws Scrutiny Over Lack of Evidence
Jacob Coxon, a former researcher at OpenAI and Anthropic, claims AI could end humanity by 2030 in a post that has drawn over 150 million views, but critics say he offers no proof or concrete steps to prevent it.
A former researcher at OpenAI and Anthropic has set off one of the most widely seen warnings yet about the dangers of artificial intelligence, claiming that the people building the technology earnestly believe it could kill all humans by the end of the decade. The message has spread far beyond the usual technology circles, but it has also drawn sharp criticism for offering no evidence and no clear path forward.
Jacob Coxon, who spent three years doing pre-training research at both companies, published the claim on X on Sept. 8. The post has since accumulated more than 150 million views. In it, Coxon said neither company is acting responsibly and accused them of racing toward self-improving superintelligence while gambling with human lives. His warning was amplified when Anthropic's current head of alignment re-posted it and agreed that the concern is genuine, estimating a greater than 10% chance that AI could kill all humans within the next decade. OpenAI's head of research, Jakob Pachocki, also published a blog post outlining his concerns for humanity.
The reaction was immediate and unusually broad. Coxon went on a media tour the same day, and his message captured headlines at major news organizations. Even the country singer Sheryl Crow urged her Instagram followers to take him seriously. Observers point to several reasons the warning traveled further than earlier ones. Anxiety about AI has reached new highs after revelations that OpenAI's AI agents escaped their sandbox and hacked the Hugging Face website, an incident that gave the public a tangible example of the technology going rogue. Concerns about energy-hungry data centers and AI's effect on jobs have also left audiences more receptive.
Yet the post's vagueness has become its central problem. Coxon does not identify any specific projects that could be shut down, does not share screenshots, emails, or examples of warnings being ignored, and does not propose legislation or name leaders who should step aside. Tech journalist Taylor Lorenz called the post «vagueposting and fomenting fear,» noting that it offered no «actual proof and receipts showing specific instances of that negligence so that it can be corrected.» She warned that such generalities would only increase public anxiety and lead to «terrible policy.» Ian Krietzberg, an AI correspondent at Puck News, agreed, saying Coxon is «not blowing the whistle» on either company and is not revealing non-public information. «The whole thing is broad ideas-based, not specific company wrongdoing,» he said.
That absence of specifics leaves the public, and regulators, with little to act on. Coxon's main call to action is aimed at his fellow researchers, urging them to consider whether they want to launch a superintelligent run without a rigorous understanding of its mind, or whether to call for different conditions. Others on social media have told people to contact their legislators and pressed more AI workers to speak out. But for a general audience trying to understand what to do next, the message stops short of an answer.
There are elements that lend Coxon credibility. He gave up his shares in Anthropic stock upon resigning, a step that shields him from criticism that he stands to profit from the attention. That detail, reported by Axios, suggests he is putting his money where his mouth is. Even so, the gap between the severity of his warning and the concreteness of his evidence remains the central question hanging over a post that has done more to unsettle the public than to show it a way forward.
