A British engineer from Anthropic who for the last three years has also worked for OpenAI, the company that created the famous ChatGPT, has resigned, accusing both companies of “not acting responsibly” and of “playing with our lives”.
Jacob Coxon, a software engineer listed in the media as one of the developers of GPT-4o and GPT-4.5 (although without management roles), stated in a series of posts on his
The twenty-seven-year-old engineer believes that the power of this technology is underestimated; claims that AI projects “will soon become superhuman systems capable of hacking anything, revolutionizing every industry, and acquiring real power and resources.”
In his opinion, both OpenAI and Anthropic are aware of this reality, but there is a difference: the former “may not have fully internalized the challenges that put civilization at risk”, while the latter understands them well but is “trapped in the race to be first” even if, in the most optimistic scenario, the goal is to be first “precisely to be able to act responsibly”.
Coxon sees no solution other than coordination between the various AI projects and the companies involved, yet he believes this is not happening. He therefore argues for the need for “challenging actions, such as a temporary suspension of model capacity building”, and invites researchers in the sector to think deeply about the issue.
“You want to power a super-intelligent “reinforcement learning” (RL, in the jargon) system without a rigorous understanding of its mind?” he asks colleagues who deal with AI «Will you continue to remain silent because “a lot is happening anyway”, or will you seize this moment to support different conditions?».