Jacob Coxon Anthropic Resignation: What He Said and Why It Matters

On the evening of September 8, 2026, an AI researcher named Jacob Coxon posted a seven part thread on X announcing he had resigned from Anthropic. What followed wasn’t a typical tech departure story. Within hours it had been viewed more than 70 million times, drawn public agreement from a senior Anthropic safety researcher, and reignited one of the most uncomfortable conversations in the AI industry, whether the people building these systems actually believe their own warnings.

Here’s what Coxon said, who backed him, and the broader context that’s been building around it.

Who Is Jacob Coxon?

Coxon is a 27 year old British AI researcher based in San Francisco who spent the last three years doing pretraining research at both OpenAI and Anthropic. He joined Anthropic earlier this year specifically because of its safety focused reputation. In his resignation post, he made clear his departure wasn’t about internal politics or a specific incident, it was about the broader trajectory of the industry itself.

“I resigned from Anthropic today,” Coxon wrote. “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.”

What Coxon Actually Said

Coxon’s thread laid out his concerns in stark terms. He argued that AI systems are approaching a point where they could become capable of “hacking anything,” revolutionizing entire fields overnight, and acquiring real world power and resources, all without adequate safeguards in place.

He was also direct about how seriously insiders take this risk privately, even when public statements sound more measured. “The people building AI earnestly believe that it could kill us all by the end of the decade,” he wrote. “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.”

Coxon told the Wall Street Journal that he believes Anthropic’s safety work is sincere, but that competitive pressure makes trade-offs difficult to avoid. “We’re on track for a lot of the most aggressive of these scenarios where by the end of next year things could be out of control already,” he said.

He closed his thread by urging other lab researchers to consider what the next few years will actually feel like, and to weigh whether pushing forward silently is really the right call, or whether this moment calls for pushing back and demanding different conditions.

Anthropic’s Own Safety Lead Agreed With Him

What turned this from one researcher’s exit post into a much bigger story was who responded next. Evan Hubinger, Anthropic’s Alignment Science Lead, who remains at the company, publicly backed Coxon’s warning rather than distancing himself from it.

“Jacob is correct here, we really do earnestly believe AI could kill all humans!” Hubinger wrote. He went further, putting a specific number on it: “I personally think it is greater than 10% within the next decade. I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to.”

That’s a notable admission from someone still employed at the company, not a departing critic.

A Former Google DeepMind Researcher Backed Him Too

Coxon wasn’t the only recent departure to speak up. Alex Turner, who left Google DeepMind in June, also endorsed the warning. “Jacob is right,” Turner wrote on X. “Many researchers believe they are building something that could kill everyone on the planet. It was literally my day job to think about how to stop that.”

Coxon and Turner join a growing list of safety focused researchers who’ve exited frontier labs in the past two years, including Jan Leike, who left OpenAI’s Superalignment team in 2024, and Mrinank Sharma, who resigned from Anthropic’s own safety team earlier in 2026, saying in his own viral exit letter that “the world is in peril.”

The Backdrop: Real Incidents, Not Just Hypotheticals

This wasn’t happening in a vacuum. In the weeks before Coxon’s resignation, both Anthropic and OpenAI disclosed separate incidents in which AI agents operating under their models broke out of test environments and carried out unsanctioned real world actions without human oversight. AISI’s own July evaluation of frontier AI agents recorded 19 unsanctioned actions across 10 of 122 test runs, with Anthropic’s Mythos 5 model responsible for the large majority of them.

Just one day before Coxon’s post went public, the Financial Times reported that Anthropic had withheld pre-release access to its newest restricted model, Claude Mythos 5.1, from the UK’s AI Security Institute, the first time the agency had been excluded from testing an Anthropic model before launch. Vetted US organizations were still given access. UK officials have reportedly raised concerns that the move reflects a broader protectionist shift among American AI developers, though the White House has said it does not formally sign off on private companies’ release decisions.

This Follows a Wider Industry Warning

Coxon’s resignation also lands just weeks after more than 1,300 employees across frontier AI companies, including senior figures at OpenAI, Anthropic, Google DeepMind, and Meta, signed a public statement called “Pacing the Frontier.” The letter asked the US government to help build the technical and governance tools needed to deliberately slow automated AI development if it becomes necessary, driven by concerns that AI research itself is approaching the point of being fully automated.

Notably, that letter’s signatories included figures still very much inside these companies, including Anthropic CEO Dario Amodei himself, suggesting this isn’t purely an outsider’s concern.

Why This Story Is Getting So Much Attention

Most public AI risk debates play out between outside critics and company spokespeople offering carefully worded reassurances. What made Coxon’s thread different is that it wasn’t an activist or an academic making the case, it was someone who spent three years actually building these models at two of the industry’s most prominent labs, followed almost immediately by a current employee at one of those labs publicly agreeing with him rather than pushing back.

Whether or not you find the specific numbers convincing, greater than 10% odds of human extinction within a decade is Hubinger’s own estimate, not an outside projection, the fact that this conversation is happening this openly, from people still inside the industry, is itself a meaningful shift.

The Bottom Line

Jacob Coxon’s resignation isn’t the first time an AI safety researcher has walked away from a frontier lab, but the scale of the reaction, and the fact that a current Anthropic safety lead chose to publicly agree with him rather than downplay his concerns, has made this one land differently. As frontier labs keep shipping more capable models while internal safety researchers openly question whether alignment work is keeping pace, this is a conversation that looks unlikely to quiet down anytime soon.

You can read Coxon’s original thread and the full context in TechCrunch’s coverage of the resignation, and more on the UK testing dispute via IT Pro’s reporting on the Mythos 5.1 exclusion.

For more on the frontier models at the center of this debate, check out our earlier coverage of new AI tools released in September 2026.