BREAKING: Anthropic researcher Jacob Coxon resigns, warning the company is racing toward superintelligence that “could kill us all by the end of the decade.”
Jacob Coxon spent three years on pretraining at
and
. Today he resigned. His claim is specific. Both labs, he says, are racing toward self-improving superintelligence without the control to match the capability. The people building the systems privately treat the downside as civilizational. Publicly they soften the language. The distinction he draws between the two companies is very important. At OpenAI, many have not internalized the stakes. At Anthropic, the stakes are understood, and the company proceeds anyway, because it assumes no one else will act responsibly. That is the logic of a race: the safest lab becomes the one that cannot afford to slow down. Coxon is not calling for an end to AI. He is calling for different conditions. Warning events such as the Hugging Face attack, he argues, have made pacing agreements among U.S. labs more plausible. A global race, in his view, may still require costly steps, including a temporary halt on capability jumps. The architecture here is familiar. Frontier models are being built inside private companies, on private timelines, with national-security and civilizational consequences. The decision to enter what some internally call the “endgame” is being taken in Slack channels, not in a process designed for that scale of risk.
Made with Grok Imagine·Make your own
Quote
Jacob Coxon
@hilbertspaess
·
12h
I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.