Jacob Coxon, a 27-year-old British software engineer who worked at Anthropic and for the past three years at OpenAI, recently resigned claiming that both companies "are not acting responsibly" regarding artificial intelligence. Coxon, identified by the media as one of the developers of GPT-4o and GPT-4.5, warned in a series of posts on social media X that the people behind the exponential development of AI "believe it could kill us all by the end of this decade", considering that "there is no other human activity that represents this level of threat".
The engineer argues that the power of the technology is being underestimated, as AI projects "will soon become superhuman systems capable of hacking anything, revolutionizing all areas and gaining real power and resources". For Coxon, both OpenAI and Anthropic are aware of this reality, but he points out a difference between them: the first "may not have deeply internalized the civilizational challenges", while the second understands them well, but "is trapped in the race to get there first".
Coxon sees no other way out besides coordination between the different AI projects and the companies involved, but feels this is not happening. Therefore, he advocates for "high-cost measures, such as a temporary ban on improving model capabilities" and asked his colleagues: "Do you want to drive super-intelligent reinforcement learning without a rigorous understanding of your own mind?"
In response, Evan Hubinger, a researcher at Anthropic, partially agreed with Coxon, stating that "we genuinely and sincerely believe that AI could kill all humans" and that personally he considers the probability greater than 10% in the next decade. Coxon is not the only researcher to distance himself from AI model development: Geoffrey Hinton, Nobel Prize in Physics winner, left his position at Google in 2023 so he could criticize AI risks without corporate restrictions, and in 2024 former OpenAI employees expressed concern about the rapid development of the technology and the absence of laws protecting whistleblowers.
