Jacob Coxon Warns of Human Extinction and Triggers a Preference Cascade

Jacob Coxon has become the latest AI insider to say the quiet part out loud: he quit Anthropic after three years on pretraining work at both Anthropic and OpenAI, and says neither lab is acting responsibly. His claim is not subtle. He says the companies are racing toward self-improving superintelligence while gambling with everyone’s lives, and he warns that these systems will soon be able to hack broadly, reshape fields overnight, and gather real power and resources.

What makes the moment different is not just the warning itself, but who is now willing to say it publicly. Coxon’s post helped trigger what the source calls a preference cascade, with more employees and former employees at major labs speaking up in public rather than only in private. The list is long: Anthropic’s Evan Hubinger, Drake Thomas, Samuel Marks and others; OpenAI staff including Roon and several more; plus people tied to Google DeepMind. The tone ranges from measured to panicked, but the direction is the same. These are people who work on the systems, and many of them say the danger is not abstract.

The public version of this fear is finally getting picked up by mainstream outlets. The Wall Street Journal led with Coxon’s resignation over out-of-control AI fears, and CNN, the FT, Axios, Fortune, Fox Business and the BBC all followed with variations on the same theme: AI employees warning that their own industry may be gambling with human survival. That is the surprising bit here. This is no longer just a safety crowd talking to itself. The alarms are coming from inside the labs that are building the thing.

Some of the quotes are blunt enough to sound fictional if they weren’t real. Coxon says the people building AI believe it could kill everyone by the end of the decade. Hubinger says Anthropic does not yet have a plan to solve alignment for superintelligence. Samuel Marks says many developers desperately want to slow down. Roon at OpenAI, while more cautious on the numbers, still argues that human extinction from machine intelligence is a real possibility and that governments and companies need to take it seriously. The basic picture is ugly and, for now, oddly consistent: the people closest to the work think progress is moving faster than the safety plan.

And the timing matters. The source says events over the last two months, and especially the last week, have made a lot of people more willing to speak. That is how these cascades work. One person quits, another posts, a third gives a TV interview, and suddenly the story is no longer a fringe warning but a growing internal revolt.