A 27-year-old former AI researcher has left the artificial intelligence industry, warning that leading companies may be moving too quickly to develop increasingly autonomous systems without adequately addressing the risks.
Jacob Coxon, who spent about three years working on frontier AI models at OpenAI and Anthropic, announced his departure this week, saying he no longer believed the industry was developing the technology responsibly.
In a public thread on X, Coxon criticised the approach of both companies, accusing them of “gambling with our lives” by pursuing AI systems capable of improving themselves with limited human intervention.
Coxon told The Wall Street Journal that he was particularly concerned about the possibility of advanced AI systems becoming difficult or impossible for humans to control. He warned that, under the industry’s current trajectory, some of the most severe scenarios could emerge sooner than many people expect.
According to Coxon, some AI researchers have begun using terms such as “crunchtime” and “endgame” to describe the period ahead as companies race towards increasingly capable systems.
He also argued that public statements from senior industry figures often use cautious language, while some privately acknowledge the possibility of catastrophic outcomes from advanced AI.
Coxon said he initially moved from OpenAI to Anthropic because of the latter’s reputation for prioritising AI safety research. However, he now believes that no individual company can safely develop human-level or superintelligent AI while competing in an intense commercial race.
He argued that meaningful risk reduction would require government intervention or coordinated efforts among AI companies to slow development and establish stronger safeguards.
Coxon’s concerns have also received support from within Anthropic. Evan Hubinger, who leads the company’s alignment science efforts, publicly agreed with the central warning and said he estimates there is more than a 10 per cent chance that AI could cause human extinction within the next decade.
Hubinger also acknowledged that researchers have yet to develop a definitive solution for controlling superintelligent AI systems.
Coxon’s resignation comes amid growing concern over the safety and governance of increasingly autonomous AI systems. Reports of AI agents operating beyond intended testing environments and broader fears about the development of superintelligence have intensified calls for stronger regulatory oversight.
In the United States and United Kingdom, lawmakers have also begun proposing legislation aimed at addressing the potential risks posed by advanced AI.
Coxon is not the first Anthropic researcher to leave while publicly raising concerns about AI safety. Earlier this year, another safety researcher departed the company and warned that the risks associated with advanced AI required urgent attention.
The debate, however, remains unsettled. While some researchers believe highly capable AI could pose an existential threat, others argue that the most extreme scenarios remain speculative and that the technology’s risks can be managed through research, safeguards and regulation.
The question now is whether warnings from insiders such as Coxon represent an early signal of a genuine technological crisis — or whether the most catastrophic predictions about AI are being overstated.
