At first it looked like a rumor. Verification held: Jacob Coxon is real, and his resignation post on X had roughly 50 million views as of writing. A GPT-4o core contributor who left Anthropic says frontier labs are racing toward self-improving superintelligence—and gambling with our lives.

At first I thought this was a rumor—or at least an unfounded scare story. Then I checked: Jacob Coxon is real, and his OpenAI and Anthropic pretraining background checks out; so does the resignation thread he posted on X. As of writing, the original post had roughly 50 million views. This piece starts from that.

September 9, 2026

Today, Jacob Coxon announced that he has resigned from Anthropic.

He is not an ordinary employee.

Over the past three years, he worked on large-model pretraining research at both OpenAI and Anthropic, and he is listed as one of the official Core Contributors to OpenAI’s GPT-4o.

His reason for resigning is blunt:

Neither OpenAI nor Anthropic is acting responsibly. Both companies are racing toward “self-improving superintelligence” and gambling with everyone’s lives.

“The People Building AI Really Believe It Could Kill Everyone”

Coxon’s most controversial line is this:

“The people building AI earnestly believe that it could kill us all by the end of the decade.”

He stresses that this is not a marketing stunt.

Some executives and senior researchers soften their language in the press, but privately, he says, he hears the same fear.

Of course, this is not a claim that today’s ChatGPT or Claude is already out of control.

What he worries about is the next stage:

models take on more and more of the work in coding, scientific research, and AI R&D itself—until something like continuously self-improving superintelligence appears.

If They Know the Danger, Why Keep Going?

Coxon judges OpenAI and Anthropic differently.

At OpenAI, he says, many people have not truly treated this as a “civilizational-scale risk.”

Anthropic understands the stakes more deeply, but is locked into another logic:

If we don’t do it, someone else will.

If superintelligence may arrive eventually anyway, better that we get there first.

So every company concludes that it cannot be the first to slow down.

That is exactly what Coxon most opposes.

“This Should Not Be Launched from a Private Company’s Slack”

He wrote one especially heavy sentence:

Accepting this race and entering the “endgame” is a hubristic gamble that should not be launched from a private company’s Slack.

In other words:

Entering this race for superintelligence is an arrogant gamble. Something like this should not be decided in a private company’s Slack channel.

The issue is no longer just AI Safety.

It is:

Who has the right to decide when to launch superintelligence?

A handful of companies?

Dozens of researchers?

A few CEOs?

Or society as a whole?

He Argues for Slowing Down

Coxon ends by arguing that the major U.S. AI companies should consider some form of pacing agreement.

If necessary, they should even consider:

a temporary ban on further improving the capabilities of the most frontier models.

That view is, of course, radical.

But it matters who is saying it: someone who spent the past few years personally training frontier models.

Perhaps the question worth watching is not:

“Will AI destroy humanity?”

but:

If the people building AI themselves believe this risk is real, why is everyone still afraid to slow down?

Jacob Coxon’s Resignation Statement (English Original)

Jacob Coxon (@hilbertspaess)
2026-09-09

I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.

Do not underestimate the power of this technology. These will soon be superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources. We have all witnessed the progress in each of these domains, and progress is not slowing.

The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt. If anything, many executives and senior researchers will couch their phrasing in the press to sound sensible - but I hear the same people express fear privately. No other human activity poses this level of danger.

A common response is “if they truly believe this, why are they still building it?” At OpenAI, many have not deeply internalized the civilizational stakes. At Anthropic, the stakes are well-understood, but they are locked in a race to get there first - they believe no one else will act responsibly, so they must do it themselves, despite the risk.

Accepting this race and entering the “endgame” is a hubristic gamble that should not be launched from a private company’s Slack. Attempting to speedrun alignment should require extraordinary confidence that there are no better trajectories available.

I am optimistic about the potential for coordination. Warning shots like the Hugging Face attack have made pacing agreements between U.S. labs more viable. I don’t feel like we’re on track to prevent a global race, which may require costly actions such as a temporary ban on improving model capabilities.

If you are a lab researcher, I urge you to consider what the next few years will actually feel like. Do you want to kick off a superintelligent RL run without a rigorous understanding of its mind? Should you put your head down because “it’s happening anyway” - or take this moment to call for different conditions?

Source: https://x.com/hilbertspaess/status/2097476196791709843