Anthropic researcher resigns, warns AI could lose control by 2027

Huo Xing Cai Jing reports that Anthropic researcher Jacob Coxon has announced his resignation and plans to leave the AI industry. Over the past three years, he has conducted pretraining research at both OpenAI and Anthropic. He stated that neither company is responsibly advancing AI, but instead is racing to build superintelligences capable of further enhancing their own capabilities—“betting our lives on it.” He moved from OpenAI to Anthropic this year precisely because Anthropic appeared to prioritize model safety more seriously. Coxon acknowledges that Anthropic’s safety efforts are genuine, but argues that as long as companies continue competing to be the first to achieve AGI (artificial general intelligence capable of outperforming humans across a wide range of tasks), compromises on safety will remain inevitable, and no single company’s self-regulation can solve the problem. His assessment of the two companies is this: Many at OpenAI still do not fully grasp the magnitude of the risk; Anthropic understands the risk but believes others cannot be trusted, so it feels compelled to be first. Coxon says that some AI professionals are privately far more pessimistic than their public statements suggest, with some even fearing AI could lead to human extinction before the end of this decade. He personally judges that, under the most aggressive scenario, AI could become uncontrollable by this time next year—and therefore, such decisions should not be made solely by engineers within a few companies.