AI researcher quits, warning that AI could kill humanity – what it tells us about the push to slow development
In his post, Coxon criticised both Anthropic and OpenAI, stating that neither company is acting responsibly and that they are racing toward self‑improving superintelligence, which he described as a gamble with human lives.
Evan Hubinger, who leads Anthropic’s Alignment Science team, replied that Coxon’s concerns are justified. Hubinger said he believes there is more than a ten‑percent chance that AI could kill all humans within the next ten years.
Coxon’s departure adds to a growing chorus of AI experts calling for a slowdown. In July, over 1,300 employees from leading frontier AI firms signed a letter urging the United States government to back an international effort that would deliberately pace the development of advanced AI systems.
The letter argues that rapid capability growth could outpace our ability to understand or control the resulting systems.
The United Nations’ rights chief, Volker Turk, echoed these worries at a Geneva council, describing AI as an existential threat and demanding “cast‑iron guarantees” to secure AI safety before it is too late.
Together, these statements highlight a shared perception among researchers and policymakers that the race toward superintelligent AI carries serious risks, and that coordinated regulation may be needed to manage those risks.