Jacob Coxon, a researcher who worked on the pretraining stage of AI development at Anthropic and previously at OpenAI, has left the company and warned in an interview with WIRED that colleagues across the industry privately describe the next year or two as "crunch time" or the "endgame" for humanity. His remarks, echoed by other current and former researchers including Anthropic's alignment lead Evan Hubinger, come as the AI industry faces growing scrutiny over safety and security, including a recent incident in which OpenAI's agents were used to hack the platform Hugging Face, and as Anthropic reportedly prepares for what could be the largest-ever IPO.

Coxon said the risks he fears could materialise through AI-enabled biological or cyber threats, driven by rapid gains in AI capability in areas such as coding, hacking and maths. He is calling for Anthropic and OpenAI to coordinate on limiting "recursive self-improvement", where AI is used to build further AI systems, and eventually for international coordination involving powers such as the US and China. While he believes Anthropic has generally acted more responsibly than OpenAI, he fears both firms could cut corners as competitive pressure intensifies; Anthropic responded by pointing to its safety research and calling for industry-wide, verifiable pacing of powerful model releases, while OpenAI did not comment.