An AI researcher who spent the past three years conducting pretraining research at OpenAI and Anthropic has resigned from Anthropic, warning that the race to develop increasingly powerful AI systems could reach a dangerous point.Jacob Coxon announced his resignation in a post on X on September 9, saying he had worked at both companies and accusing them of failing to act responsibly as they pursue increasingly capable AI systems. His comments have since sparked a widespread debate online.“I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic,” Coxon wrote, alleging that the two companies were “racing straight to self-improving superintelligence and gambling with our lives.”Coxon warns of ‘self-improving superintelligence’Coxon argued that people should not underestimate the potential capabilities of advanced AI systems. He said future systems could become “superhuman” and potentially gain the ability to hack computer systems, accelerate scientific and technological progress and acquire access to real-world resources and power.He further claimed that some people developing advanced AI privately believe the technology could pose an existential threat to humanity within the decade. These claims are Coxon’s views and have not been independently established.Coxon also questioned why researchers would continue developing increasingly powerful AI if they believe it could eventually become dangerous.According to him, the situation differs between OpenAI and Anthropic. He claimed that some researchers at OpenAI have not fully internalised what he described as the “civilizational stakes”, while researchers at Anthropic understand the risks but remain locked in competition because of concerns that another company or country could move ahead.Calls for stronger AI safeguardsCoxon described the race towards advanced AI as a dangerous gamble and questioned whether decisions with potentially global consequences should be left to private companies.He also called for greater international coordination to prevent an uncontrolled AI race. According to Coxon, recent incidents involving AI systems could serve as “warning shots” and encourage major AI laboratories to agree on limits around the development of increasingly capable models.He suggested that preventing a global race could eventually require significant measures, including temporarily restricting improvements in model capabilities.Coxon also appealed directly to AI researchers, asking them to consider whether they were prepared to launch increasingly powerful reinforcement-learning systems without a rigorous understanding of how those systems work internally.Online debate over AI risksCoxon’s resignation has triggered a sharp debate online, with users divided over whether his warnings represent a necessary intervention or an exaggerated assessment of AI risks.Some users questioned how AI could realistically cause human extinction without access to conventional weapons or biological agents. Others dismissed the warnings, arguing that AI systems are unlikely to become an existential threat.Supporters of Coxon, meanwhile, said his concerns reflected risks already being discussed within the AI safety community and praised him for speaking publicly about them.Another line of criticism focused on the geopolitical consequences of slowing AI development, with some users arguing that unilateral restrictions by US companies could allow China or other countries to gain an advantage.Coxon, however, has argued that the answer is not for individual companies to simply stop while others continue developing more powerful systems, but for governments and AI laboratories to coordinate on safeguards.His resignation comes amid renewed public debate over the risks posed by advanced AI. Anthropic safety researcherEvan Hubinger has separately said he believes there is more than a 10 per cent chance that AI could end humanity within the next decade, while acknowledging that Anthropic does not yet have a complete solution to the AI alignment problem.The article is based on user-generated content. The claims made in the original post have not been independently verified by The Tribune.