On September 9, a 27-year-old British AI researcher named Jacob Coxon published a resignation statement on social platform X, sending shockwaves through the entire AI industry. Coxon announced his departure from Anthropic and issued a public warning that both OpenAI and Anthropic are "betting with the lives of all humanity" as they race to develop self-improving superintelligence without acting responsibly. The statement amassed over 100 million views within a short period, becoming one of the most talked-about events in the AI sector this week.
According to Coxon's post on X, he spent the past three years working on large model pre-training research, first at OpenAI and later at Anthropic. He joined OpenAI's technical team in 2023 and moved to Anthropic as a researcher in July 2026. During his time at OpenAI, he contributed to the development of GPT-4o and was listed as a core contributor to the model's system card. However, after only about two months at Anthropic, he decided to leave and announced his exit from the AI industry altogether.
In his resignation statement, Coxon used unusually harsh language, writing that both companies' actions are irresponsible as they compete to build self-improving superintelligence, wagering our lives in the process. He further warned that these AI systems will soon possess "superhuman" capabilities, enabling them to conduct hacking operations, reshape entire industries overnight, and seize real-world power and resources. Coxon also shed light on a deep contradiction within the industry: AI practitioners privately genuinely believe that AI "could kill us all by the end of the century," yet many executives and senior researchers soften their rhetoric to appear rational in front of the media.
He drew a distinction between the internal cultures of the two companies. At OpenAI, he claimed, many employees do not truly recognize that this race concerns the fate of human civilization. At Anthropic, meanwhile, people are aware of the risks but have fallen into a race to be first, operating under the logic that since no one else will act responsibly, they must achieve superintelligence themselves even at the risk of safety. Coxon also specifically pointed out that within Anthropic, critical decisions are often made only in private company Slack channels, which he argued should not be the case. He called on leading US AI laboratories to establish protocols for capability development pacing and, if necessary, to temporarily halt further model capability improvements.
Musk Steps In and Questions the Impact
The statement was initially reposted by X user @XFreeze, who publicly questioned whether anyone at Anthropic could confirm that Coxon was indeed one of their employees. Elon Musk promptly replied beneath that post, confirming Coxon's work history but noting that he did work there, albeit only for a few months. However, as Coxon's post continued to rack up engagement metrics, with one tracker counting over 123 million views, more than 656,000 likes, and over 190,000 new followers, Musk expressed skepticism about the authenticity of these numbers, saying he did not believe a new account with virtually no prior posting history could generate such high levels of interaction.
The controversy surrounding Coxon did not stop there. Investigative researchers pointed out that Coxon's post was published just 18 minutes before a Wall Street Journal exclusive report citing his statement, suggesting coordination with the media in advance rather than a spontaneous decision. Additionally, the three accounts that first reposted Coxon's message belonged to three AI policy advocacy organizations: Encode AI, the AI Policy Network, and the AI Futures Project. Their funding was traced back to the Survival and Flourishing Fund, backed by Skype co-founder and Anthropic investor Jaan Tallinn.
Anthropic Safety Lead Publicly Endorses Coxon's Claims
Within Anthropic, an unusual public reaction emerged. Evan Hubinger, Anthropic's alignment science lead, responded on X in a personal capacity, stating that "Jacob is right." He agreed with Coxon's core assessment and said he personally believes the probability of AI causing human extinction within the next decade exceeds 10 percent. Hubinger also acknowledged that Anthropic currently has no clear solution to the superintelligence alignment problem and cannot be certain the company is headed in the right direction. He noted that Anthropic's risk assessment for existing models remains low, but the real concern lies in the possibility that AI participating in the development of next-generation AI could lead to increasingly rapid capability iteration.
Anthropic's safety policies have also evolved in response to the competitive landscape. In 2023, the company introduced its "Responsible Scaling Policy," which committed to not training or deploying models that breach risk thresholds without adequate safety measures in place. That red line once served as a key differentiator between Anthropic and other major model companies. However, in February 2026, Anthropic significantly revised that policy. The updated version retains risk reporting, model evaluations, and external reviews, but no longer unconditionally commits to halting training when safeguards are insufficient. Anthropic's chief scientist, Jared Kaplan, said at the time that unilaterally stopping training would not make the world safer if competitors continued to push forward.
Meanwhile, Anthropic is preparing for its initial public offering in mid-October. Against this backdrop, the public resignation of an internal researcher and the public endorsement from a safety team lead undoubtedly cast a shadow over the company's listing prospects.