A researcher who worked on AI safety at Anthropic just quit, and used his exit to warn that the company's own products might end civilization.
Jacob Coxon left Anthropic and laid out his reasoning in a social media thread posted Tuesday night. He argued the real danger isn't today's chatbots but the coming wave of "self-improving superintelligence": systems capable of hacking anything, upending entire industries overnight, and grabbing real-world power and resources. Coxon said frontier AI companies are "gambling with our lives," believing their own creations could kill everyone "by the end of the decade." He added that colleagues either haven't grasped the stakes or think they must "speedrun" toward superintelligence before a less careful competitor gets there first.
This isn't a fringe complaint from someone with an axe to grind. Evan Hubinger, who leads Anthropic's Alignment Science team, publicly backed Coxon up, saying he personally puts the odds of AI killing everyone at more than 10 percent within the next decade. That's a number from a company insider, not an outside critic, and it means the people building these systems are, by their own account, racing toward technology they believe could kill everyone on the planet.
Anthropic has built its brand on being the safety-conscious AI lab. A safety lead handing out double-digit odds of extinction suggests the marketing and the internal math don't quite agree.