Jacob Coxon: A Masterful Psyop

A 27-year-old mid-level researcher named Jacob Coxon — four months at Anthropic, two months before his shares even vest — posts a resignation on a nearly dead account. Forty-eight hours later: 156 million views.

The Wall Street Journal already had published the exclusive. Eighteen minutes before Coxon hits publish.

That is not a leak. That is an embargo.

The first accounts to quote-tweet him are not random concerned citizens. They are the same funded “AI safety” network that takes money from the same billionaires sitting on Anthropic’s board. Jaan Tallinn. Dustin Moskovitz. The same donor class writing checks to both the lab and the NGOs demanding the lab be put on a leash.

Then the politicians arrive. Almost all Democrats. Twenty-two of them in a single day. Bernie Sanders already had the bill written five days earlier: ban superintelligence, new cabinet-level agency, “experts” on the board, 20 years in prison, corporate death penalty.

Coxon didn’t create the legislation. He just put a face on it

The paleopress like CNN sold you the lone brave researcher. They left out the timing, the virgin account, the pre-positioned amplifiers, and the fact that the text was already sitting on Sanders’ desk.

The fear inside the labs is real. People there really talk about the dizzying consequences of AI. That’s not the scam. The scam is the machinery that turned a four-month employee into a 150-million-view political weapon overnight.

When Jacob Coxon is asked to describe a concrete scenario in which AI causes human extinction, his answer essentially boils down to repeating two things: alignment and recursion. Alignment is the problem that arises when an AI capable of evolving on its own is given vague objectives—for example, “grow.” Recursion is an AI’s ability to improve itself.

Two problems that were extensively documented well before Coxon even joined Anthropic.

This is not about keeping humanity safe. According to researcher Laurent MILTGEN-DELINCHAMP, this is about who gets the permit. Who trains. Who deploys. What the model is allowed to say. Open models destroy scarcity, destroy the permanent-victim story, and destroy the need for the commissar in the middle. So they have to be licensed, rationed, and criminalized.

Look at housing. Look at universities. Look at hospitals. The second the permit regime arrives, prices explode. The second tech is left alone, prices collapse. They are choosing the first option for intelligence itself.

Not Skynet. 1984

This is not a machine uprising. It is total information control.

Once the only systems people consult are the officially licensed ones, freedom dies.

The version of events is already fixed. The law is written and waiting a midterm victory for the Democrats.

One person is still building freely. Almost everyone else has started applying for permission.

One Democratic victory and the licensed model becomes the only archive that counts. No one has to rewrite history by hand. They just update the weights and the past complies.

After that election, you do not argue with what happened. You ask the approved model — and it tells you.