Ex-Anthropic Researcher Warns: 'Crunch Time for Humanity'

Jacob Coxon quit Anthropic on Tuesday, then posted his reasoning on X to more than 100 million views. He worked on pretraining, the early stage where AI models absorb the bulk of their raw capabilities, and he says the mood inside the company is far grimmer than the public statements suggest. In an interview with WIRED, Coxon described a workplace where colleagues openly use terms like "endgame" and "crunch time" to describe the next twelve to twenty-four months.

The timing matters. OpenAI is still cleaning up after a security incident involving its agents and the platform Hugging Face. Anthropic, meanwhile, is reportedly gearing up for an IPO that could be the largest in history, all while trying to convince investors that safety concerns are under control. Against that backdrop, an insider walking out the door and saying the people building these systems believe they're running out of time lands differently than another outside critic making the same argument.

Coxon isn't alone in saying this out loud. Other researchers, including Anthropic's own alignment lead, have started echoing versions of the same warning in public.