Jacob Coxon spent three years training AI models. First at OpenAI, then at Anthropic. On September 8, he quit and went public.
"They are racing straight to self-improving superintelligence and gambling with our lives," he wrote on X.
The people inside believe it
Coxon said the engineers and executives building frontier AI earnestly think it could kill everyone by the end of the decade. This isn't marketing. It's what they say in private.
Evan Hubinger leads alignment science at Anthropic — the team paid to keep AI under human control. He replied publicly: "We really do earnestly believe AI could kill all humans."
Hubinger said he personally puts the risk above ten percent within ten years. He also said Anthropic has no plan to control superintelligence and isn't clearly on track to get one.
The nightmare scenario
The fear centers on recursive self-improvement: AI systems building better versions of themselves, faster than humans can understand or stop them. Once that starts, each generation arrives quicker. Labs get less time to test. Less time to catch problems.
Coxon quit two months before his equity vested — the stock payout most Silicon Valley employees wait years to collect. He walked away from the money to sound the alarm.
What this means for you
You use these tools daily. ChatGPT. Claude. Gemini. The companies building them admit they don't know how to control what comes next. And they're not slowing down because they fear a competitor will get there first.