I have spent 25 years around enterprise technology, and I do not scare easily about vendor claims or industry drama. So when I read Jacob Coxon's resignation post, I paid less attention to the prediction and more to who was making it.
Coxon is 27. He spent three years doing pretraining research, first at OpenAI and then at Anthropic, the kind of work that sits closest to what a frontier model can actually do. In September he resigned from Anthropic and posted a warning on Slack that leaked and then went viral on X: both companies, he argued, are racing toward self-improving superintelligence and gambling with human survival to win that race. That is a strong claim, and it came from an insider, not an outside critic.
What struck me more was what happened next. Evan Hubinger, who leads alignment science at Anthropic, publicly agreed with Coxon's framing instead of distancing himself from it. He put his own estimate at greater than ten percent odds of AI causing human extinction within the decade. But read what he is actually estimating: the risk from a future system capable of recursively improving itself, not the assistant answering your email today. He called present-day model risk low. That distinction is getting lost in the headlines, and it matters.
Dario Amodei's essay a few weeks earlier, "We Must Pace the Frontier," reads differently against that backdrop. He is not calling for a halt. He is arguing that labs should slow the rate at which capability outpaces the safety work meant to keep it aligned, and he backed it with a concrete first step: permanent, employee-level access for independent evaluators inside Anthropic's own training pipeline. Sam Altman said publicly that OpenAI would do the same within hours of the essay going up.
None of this means today's tools are dangerous. It means the people closest to building them are on the record saying the current pace outruns the safety work meant to contain what comes next.
I have built my career on the belief that AI is a genuine step-change in how enterprises operate, and I still believe that. But an industry watcher's job is to notice when the builders themselves start sounding the alarm, and to hold the nuance instead of collapsing it into "AI will kill us" or "nothing to see here." Neither is what was actually said. The honest version is harder and more useful: the people building this technology are telling you, plainly, that they do not yet have a plan for what comes after it stops being a tool they fully control.
Nya Akoma. Take Heart.