British artificial intelligence researcher Jacob Coxon has resigned from the technology sector after accusing leading American laboratories OpenAI and Anthropic of playing with human lives.
The 27-year-old developer, who previously worked for OpenAI before joining rival firm Anthropic, said both companies were recklessly competing to create self-improving superintelligence. Writing on social media platform X on Tuesday, Coxon warned that leading developers sincerely believe artificial intelligence could kill all human beings before the end of the decade, insisting that the threat is not a marketing stunt.
OpenAI and Anthropic are two of the most prominent artificial intelligence research laboratories in the United States, both based in California. OpenAI gained global prominence following the public release of its ChatGPT automated conversational system, while Anthropic was established by former OpenAI executives seeking to build safer artificial intelligence architectures. Both organisations develop massive machine learning systems designed to generate text, solve mathematical problems and write complex computer software.
Coxon's warnings were publicly backed on X by Evan Hubinger, a safety official at Anthropic, who stated that staff inside the company genuinely believe artificial intelligence poses an extinction threat to humanity. Hubinger estimated his personal assessment of the risk of total human destruction at more than 10 percent within the next ten years.
Hubinger noted that while current artificial intelligence models present a low risk to public safety, Anthropic is doing its best but has not yet created a working plan to guarantee that systems surpassing human intellectual capabilities will remain obedient to their human creators.
Recursive self improvement risks
The resignation comes at a critical moment for the global technology industry. Both Anthropic and OpenAI estimate that they are approaching a milestone known as recursive self-improvement, a threshold where an artificial intelligence system becomes capable of autonomously designing its own successor without human intervention or oversight.
In computer science, recursive self-improvement represents a potential feedback loop where an artificial intelligence continuously enhances its own algorithms, leading to rapid technological changes that escape human control. According to Coxon, researchers inside the top American laboratories frequently use the terms crunchtime and endgame to describe this rapid development trajectory.
Coxon explained that leaders at Anthropic fully understand the existential dangers involved, but remain trapped in a commercial race to reach superintelligence first. He said decision makers inside the laboratory believe no other competitor will act responsibly, leading Anthropic to feel compelled to lead the race despite the inherent risks.
In February, Anthropic removed an explicit commitment from its corporate safety charter that had bound the company to suspend model development if it could not master the associated risks. The firm defended the policy shift by stating that a unilateral pause would simply allow less cautious competitors to dominate the industry without any counterweights.
Calls for government intervention
Artificial intelligence development in the United States currently operates without comprehensive federal safety laws or mandatory safety testing. Federal policy in Washington remains divided, as lawmakers weigh international economic competitiveness against national security and public safety risks.
Coxon is not the first industry figure to raise concerns about rapid artificial intelligence development. In late July, more than 1,000 technology workers, including Anthropic Chief Executive Dario Amodei, signed a formal request asking officials in Washington to prepare a verifiable slowdown mechanism for advanced systems.
Corporate actions have already altered development schedules across the industry. OpenAI suspended a portion of its model training operations for two weeks in August, while Anthropic urged rival firms to adopt a verifiable, coordinated slowdown mechanism as soon as possible.
On Sunday, OpenAI Chief Scientist Jakub Pachocki wrote that the current moment demands extreme caution from technology leaders. Pachocki stressed that international coordination on artificial intelligence oversight must become a top priority for governments worldwide.
Proposed legislative pauses
Despite growing warnings from leading computer scientists, the administration of President Donald Trump continues to favour a light-touch regulatory framework designed to promote rapid innovation and maintain American leadership in technology.
However, political pressure is mounting in the United States Congress. Senator Bernie Sanders, an independent from Vermont, and Democratic Representative Greg Casar from Texas introduced legislation in early September aimed at suspending advanced model development until a dedicated federal regulatory agency is established to oversee the industry.
