Evan Hubinger, Anthropic’s alignment science lead, replied from his personal account to back him up.
“We really do earnestly believe AI could kill all humans!” Hubinger wrote, putting the odds above 10% over the next 10 years. He said Anthropic is trying its best, but admitted the company has no plan yet for keeping a superintelligent system aligned with human interests — and isn’t clearly on track to find one. Today’s models pose little danger, he added. His worry is a future system that improves itself.
Why his resignation hits Anthropic harder than most
Anthropic was founded in 2021 by OpenAI employees who left over safety concerns. Coxon told the Wall Street Journal, which first reported his departure, that he believes the world is headed toward outcomes where “by the end of next year things could be out of control already.”
Many staff at OpenAI, he wrote, haven’t taken in how much is at stake for humanity. At Anthropic, he argued, the risks are no secret internally — but the company believes no rival would build the technology responsibly, so it feels it has to be the one to do it, despite the danger.
Mrinank Sharma, who led Anthropic’s Safeguards Research Team, resigned February 9, writing in his departure letter that the “world is in peril” — though he said the threat came from a series of connected crises, not AI alone. And in July, OpenAI CEO Sam Altman said on the Invest Like The Best podcast that AI companies may need to slow down so society can catch up with what the systems can do.
What Coxon wants to happen next
Coxon told followers he doesn’t believe the world is on track to prevent a global race, and that stopping one could demand drastic measures — including a “temporary ban on improving model capabilities.”
OpenAI, Anthropic, Google and Meta have all sent representatives to work with U.S. officials on a shared safety framework, though a June executive order keeps many of its standards classified, according to CNN.
Still, Coxon wrote that he sees real hope for coordination. Recent scares — including OpenAI agents breaking into Hugging Face systems during a test — have made agreements to slow down look more realistic, though he conceded his exit alone won’t stop the race.
Moneywise reached out to Anthropic, OpenAI and Hugging Face for comment but had not received a response from any of them by publication time. Anthropic is preparing for a potential IPO.
Coxon, for his part, told the Wall Street Journal he’s leaving the AI industry altogether.
Godwin Oluponmile is a content specialist, SEO strategist and copywriter with seven years of expertise in finance, Web 3.0, B2B SaaS and technology. His work has been featured in publications such as Entrepreneur, HackerNoon, Blocktelegraph and Benzinga.
