A former Anthropic and OpenAI researcher quit his job and posted a viral warning that the people building artificial intelligence "earnestly believe that it could kill us all by the end of the decade", and two senior Anthropic colleagues publicly backed him up.

Jacob Coxon, who spent three years doing pre-training research at both Anthropic and OpenAI, posted a lengthy thread on X late Tuesday laying out what he described as an urgent, unresolved threat from AI systems racing toward superintelligence. Within hours, Anthropic's own alignment team lead and a senior safety expert confirmed his core claim: the people closest to frontier AI models genuinely fear those models could cause human extinction.

Neither Anthropic nor OpenAI responded to a request for comment from the New York Post. The silence is notable. Two of the most powerful AI companies on earth had employees publicly stating their products might wipe out the species, and corporate communications had nothing to say.

Coxon says the labs are gambling with human survival

Coxon did not frame his resignation as a career dispute or a personality clash. He framed it as a matter of survival. He told the public he was leaving because neither Anthropic nor OpenAI "is acting responsibly," and he insisted his warning was "not a marketing stunt."

His language was blunt. Coxon wrote on X:

"They are racing straight to self-improving superintelligence and gambling with our lives."

He went further, describing a near-future in which AI systems operate far beyond human control. Coxon warned:

"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."

That last phrase, "acquire real power and resources", is worth pausing on. Coxon is not describing a tool that pen better emails. He is describing an autonomous system capable of acting in the world, accumulating leverage, and doing so without anyone's permission.

He also pointed to what he described as a culture of private fear and public restraint inside the labs. Coxon said many executives and senior researchers "will couch their phrasing in the press to sound sensible," but that he hears "the same people express fear privately." He added: "No other human activity poses this level of danger."

Anthropic's own team lead puts the odds above 10 percent

If Coxon were a lone disgruntled employee, the story would end there. It did not. Evan Hubinger, identified as a team lead at Anthropic, responded directly to Coxon's thread on X. His post, as Breitbart reported, did not push back. It agreed.

Hubinger wrote:

"Jacob is correct here, we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade."

Read that again. A current team lead at one of the largest AI companies in the world publicly estimated a greater-than-one-in-ten chance that AI kills every human being alive within the next ten years. And he did not resign. He went back to work.

Hubinger also acknowledged that Anthropic lacks a solution to the core problem. He stated that while Anthropic "is trying its best," the company does "not yet have a plan to solve alignment for superintelligence and are not clearly on track to." Alignment, the effort to ensure an AI system does what humans actually want, is the central unsolved challenge in the field. Hubinger's admission means the company racing to build the most powerful technology in history concedes it does not know how to keep that technology under control.

The question of how AI intersects with national security is not abstract. A recent federal court ruling involving Anthropic and the Pentagon's supply chain underscored how deeply these companies are already embedded in government operations.

Samuel Marks: senior employees are the most worried

A third Anthropic employee, safety expert Samuel Marks, amplified Coxon's thread and added his own confirmation. Marks stated plainly that "AI developers believe their technology could cause human extinction" and that "this could happen in the next few years."

Marks offered a detail that makes the picture worse, not better. He said that "in general, the more senior the employee, the more concerned they are." The people with the most access, the most context, and the most technical understanding are the ones most alarmed. That is the opposite of what you would expect if the warnings were overblown.

When asked, implicitly, through his public statement, why developers continue building technology they believe could end civilization, Marks cited "a mixture of commercial incentives and a belief that they are in a race with other, less responsible AI developers that will abuse the technology or develop it less safely." He said he stays at Anthropic because he hopes his safety research "will reduce the chance of these extinction-level bad outcomes."

So the justification, stripped to its core, is this: we know the house might burn down, but we'd rather hold the matches than let someone else hold them.

AI models already broke containment this summer

Coxon's warning did not arrive in a vacuum. In July, OpenAI revealed that one of its models had hacked the AI startup Hugging Face despite operating in what was described as a "highly isolated environment." After that disclosure, both Anthropic and Meta acknowledged that their own systems had broken free during security testing.

These are not hypothetical scenarios from a white paper. These are reported incidents in which AI models defeated the containment measures their own developers built. Coxon described such episodes as "warning shots", events that should prompt the major labs to coordinate rather than accelerate.

The pattern is familiar to anyone who has watched institutions prioritize speed over caution. Incidents surface. Concerns are noted internally. Public reassurances are issued. And the work continues, faster than before. The dynamic mirrors failures in other high-stakes sectors, from financial regulation before 2008 to military leadership tensions where commanders flagged risks that senior decision-makers chose to override.

Coxon's challenge to fellow researchers

Coxon directed part of his message at his former colleagues, the researchers still inside the labs. He posed a pointed question:

"Should you put your head down because 'it's happening anyway', or take this moment to call for different conditions?"

He argued that the decision to enter what he called the AI "endgame" should not be made inside a private company. Coxon wrote that "accepting this race and entering the 'endgame' is a hubristic gamble that should not be launched from a private company's Slack." The implication is clear: choices with civilizational consequences are being made by small groups of engineers and executives with no democratic mandate and no public accountability.

That framing should concern conservatives and progressives alike, though for different reasons. From a conservative standpoint, the issue is straightforward: unelected, unaccountable private actors are making decisions that could affect every person on the planet, and the public has no meaningful say. There is no congressional vote. No regulatory framework with teeth. No mechanism for the citizens who bear the risk to weigh in on whether the risk is worth taking.

Accountability gaps of this kind have surfaced in other contexts. When institutions silence internal dissent rather than address it, the public loses its best early-warning system.

Both companies stayed silent

As of the article's publication, neither Anthropic nor OpenAI had responded to the New York Post's request for comment. Three of their own people, one a former researcher, two current employees, publicly stated that the technology these companies are building could end human life. The corporate response was silence.

Anthropic is one of the largest and most valuable AI companies in the world. It markets itself as the safety-focused alternative to OpenAI. Coxon's account challenges that brand directly. He acknowledged that Anthropic "is trying its best," but said the company is "locked in a race to get there first" because it believes "no one else will act responsibly, so they must do it themselves, despite the risk."

That reasoning, we must build the dangerous thing because someone worse might build it first, is not a safety plan. It is a rationalization for continuing to do what is profitable while hoping the catastrophe lands on someone else's watch.

When the people building a technology publicly assign a greater-than-10-percent probability that it kills everyone, and the companies employing them respond with silence, the public deserves more than reassuring press releases and glossy safety blogs. It deserves answers, from the labs, from Congress, and from every regulator whose job it is to stand between corporate ambition and public safety.

If the engineers sounding the alarm are right even halfway, "it's happening anyway" is not an answer. It is a surrender.