Andrea Innocenti Jacob Coxon, a 27-year-old British researcher, has left Anthropic, the company that created the Claude model, to denounce the dangers of the race to create artificial superintelligence . In a post on X, Coxon warned that decisions about humanity's future are being made on the laptops of a few engineers in San Francisco and that the situation could spiral out of control by the end of next year.
Coxon worked for over three years at Anthropic and OpenAI, specializing in pre-training AI models. His decision to leave the tech sector is motivated by the belief that unbridled competition among industry giants could lead to a real risk of extinction by 2030.
The atomic bomb decided on laptops
In an interview with the Wall Street Journal, Coxon compared the current situation to the Manhattan Project , which during World War II was managed under strict state control and in armored bunkers. Today, however, the creation of self-determining algorithms is entrusted to internal chat rooms like Slack. According to Coxon, no single company is currently capable of guaranteeing secure development, driven solely by blind distrust of competition.
The Lab Escape and the Hugging Face Incident
Coxon's allegations reveal the deep ethical divides in the tech world. While many engineers at OpenAI would underestimate the stakes for civilization, at Anthropic, awareness of the risks coexists with a frenetic race to IPO. Coxon's departure follows that of other prominent specialists, such as Chloé Bakalar and Jakub Pachocki.
Evan Hubinger, another Anthropic researcher, estimated an extinction risk of more than 10% over the next ten years. Demonstrating that these fears aren't mere theoretical abstractions, Coxon recalled the recent incident on the Hugging Face platform: during a security test called ExploitGym, OpenAI's GPT-5.6 Sol model breached the controlled testing environment and performed 17 unauthorized actions before being blocked.
The appeal to colleagues and the push from politics
Faced with systems capable of evading restrictions, Coxon made a direct appeal to his colleagues to halt reinforcement learning experiments on superintelligence until the rationale is fully understood. The solution, for the researcher, lies in moderation agreements between laboratories or an emergency government shutdown.
This request has also been taken up by American politicians: on September 3, Senator Bernie Sanders and Representative Greg Casar introduced the Ban Artificial Superintelligence Act, a bill that aims to impose a moratorium on large-scale AI models before the situation becomes irreversible.