Anthropic resignation reopens AI extinction risk debate

On September 8, researcher Jacob Coxon told the Wall Street Journal he was resigning from Anthropic because he feared the systems the company is developing could spiral out of control and destroy humanity. His post announcing the exit collected more than 100 million views inside 24 hours, Nature reports in a feature by Elizabeth Gibney that four of the researchers on our Who's Who tracker shared this week.

Coxon is not the loudest voice inside the building. Evan Hubinger, who leads alignment science at Anthropic, has said 'the risk of human extinction is >10% within the next decade.' Around 1,400 AI-sector employees have signed an open letter calling for the industry to slow down.

Not every researcher thinks the science is there. Michael Vermeer, a science and technology policy researcher at RAND Corporation in Santa Monica, California, told Nature that the debate 'involves so many untestable claims that you just end up with a conversation that is really more like faith than something scientific.' His 2025 RAND study broke the question into three routes: nuclear weapons, biotechnology, and atmospheric modification. Full nuclear extinction was found to be infeasible. Biological and atmospheric scenarios could not be ruled out, but executing either would require sustained physical action in the world that humans could probably detect and stop.

Heidy Khlaaf, chief AI scientist at the AI Now Institute in New York City, calls the extinction framing 'fear-mongering.' She is more worried about 'AI's low reliability and accuracy rates in critical environments with life-or-death consequences.' Gibney's piece notes a 2026 incident in which false AI-generated intelligence nearly prompted the US military to board a Chinese vessel.