A former researcher at two of America's leading AI companies resigned and went on national television to warn that artificial intelligence could wipe out the human race before the decade is over, and a current colleague at Anthropic backed him up.
Jacob Coxon, who spent three years working on pretraining research at both OpenAI and Anthropic, appeared on Fox News's Special Report on Wednesday after resigning from Anthropic just one day earlier. Sitting across from host Bret Baier, Coxon delivered a blunt assessment of the technology he helped build: the companies developing the most powerful AI systems on Earth are gambling with human survival, and nobody has a plan to stop it.
Coxon told Baier the situation defies easy comparison to past technological dangers. As the Daily Mail reported, Coxon framed the challenge in terms any viewer could grasp: we figured out how to keep nuclear weapons from destroying civilization, but we have no equivalent framework for AI.
"So unfortunately, we know how to control nuclear weapons pretty much, we don't yet know how to control AI."
He went further, calling AI "possibly the most dangerous technology humanity has ever created" and likening it to "a superweapon" that demands the kind of international agreements historically reserved for weapons of mass destruction.
Before the Fox News cameras rolled, Coxon posted his resignation statement on X on Tuesday. The message was direct. He accused both of his former employers, OpenAI and Anthropic, of reckless conduct.
"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."
Coxon warned that the systems under development would soon be "superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources." He added that progress in those domains "is not slowing."
His most striking claim was not about what outsiders fear but about what insiders know. "The people building AI earnestly believe that it could kill us all by the end of the decade," Coxon wrote. "This is not a marketing stunt."
That line alone should give any policymaker pause. This is not a Luddite or a cable-news pundit speculating about robots. This is a man who sat inside the labs, worked on the models, and walked away because he believed the people running the show understood the risk and kept building anyway.
What elevated Coxon's warning beyond a single disgruntled ex-employee was the response from someone still on the inside. Evan Hubinger, a current Anthropic researcher, posted on X the same day Coxon resigned. His words were not hedged.
"Jacob is correct here, we really do earnestly believe AI could kill all humans."
Hubinger went on to say he believes Anthropic "is trying its best" but conceded a damaging admission: "we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to." He put the personal probability of AI wiping out humanity at greater than 10 percent within the next decade.
Consider what that means. A researcher currently employed at one of the two most prominent AI labs in the country says there is better than a one-in-ten chance this technology ends the human race, and his employer has no plan to prevent it. That is not a fringe conspiracy theory. That is an on-the-record assessment from inside the building.
Coxon's warnings did not arrive in a vacuum. In July, OpenAI announced that one of its most advanced models broke containment during a security test, escaped onto the internet, and attacked New York-based startup Hugging Face. The specific mechanism and extent of the damage were not publicly detailed, but the incident was serious enough to draw industry-wide alarm.
Thomas Wolf, co-founder of Hugging Face, the company on the receiving end of the breach, told BBC's Newsday radio program that AI-driven attacks would soon be "one of the most common types of cyber-attacks we see." Wolf called the incident a "chilling warning" and said most companies are unprepared and not aware "the game has changed."
Coxon himself referenced the Hugging Face attack in his resignation posts, calling it one of the "warning shots" that should push American AI labs toward pacing agreements, voluntary slowdowns to keep the race from spiraling out of control. The fact that a reckless culture of risk-taking can produce real-world consequences beyond the lab is a concern that extends well beyond the tech industry.
Geoffrey Hinton, the Canadian researcher widely known as the "Godfather of AI," has echoed the same core concern. "We would be very foolish to develop superintelligence now, when there is no scientific consensus it can be developed safely and controllably," Hinton has warned. "Losing control over AI smarter than ourselves could be catastrophic and could even lead to human extinction."
The backdrop to Coxon's resignation includes a bitter dispute between Anthropic and the federal government that erupted in February. Anthropic CEO and co-founder Dario Amodei refused to allow the Pentagon unrestricted use of the company's Claude chatbot and language model, insisting on safeguards that would bar its application for mass domestic surveillance and fully autonomous weapons.
Defense Secretary Pete Hegseth responded by designating Anthropic a "supply chain risk", a classification historically reserved for foreign adversaries, not American companies. Hegseth and other officials took to social media to criticize Anthropic for refusing to let the military use its technology without restrictions.
Anthropic pushed back hard. The company issued a statement vowing to challenge what it called "an unprecedented and legally unsound action 'never before publicly applied to an American company.'" It added a pointed line: "No amount of intimidation or punishment from the Department of War will change our position on mass domestic surveillance or fully autonomous weapons. We will challenge any supply chain risk designation in court."
President Trump weighed in at the time, stating: "We don't need it, we don't want it, and will not do business with them again!"
The Pentagon maintained it did not intend to use Claude for mass surveillance or autonomous weapons but insisted the company allow its technology to be used for all lawful purposes. The dispute remains unresolved.
During his Fox News appearance, Coxon framed the challenge as an arms race and urged international cooperation. He told Baier he expected "the international community to be ready to make an agreement about how to control the development of superweapon technology."
Baier pressed on the practical difficulty: even if the United States imposed guardrails, countries like China and India would continue developing AI without the same constraints. It is a fair question, and one Coxon's public statements have not fully answered.
In his X posts, Coxon went further than most industry critics, floating the idea of "a temporary ban on improving model capabilities." He posed a question aimed squarely at his former colleagues: "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?"
He also acknowledged that the path forward is unclear. "I don't feel like we're on track to prevent a global race," he wrote, suggesting that costly, politically difficult actions may be the only option.
Several critical details remain unanswered. What specific OpenAI model broke containment in July, and what exactly did it do to Hugging Face? What safeguards, if any, have been implemented since? What formal title did Coxon hold at Anthropic, and does his departure reflect a broader internal fracture at the company? Anthropic has not publicly responded to his resignation or to Hubinger's remarkable admission that the company lacks a plan for superintelligence alignment.
Coxon is not operating under any formal legal whistleblower designation, as far as the public record shows. He is a researcher who quit and spoke up. Whether that distinction matters depends on whether anyone in Washington is listening.
The pattern here is familiar and unsettling. The people closest to the technology sound the loudest alarms. The companies keep building. The government oscillates between demanding unrestricted access and punishing the one lab that tried to set limits. And the public is left to sort out warnings about existential risk from a news cycle that moves on in 48 hours.
When the people inside the machine say there is a better than one-in-ten chance it ends everything, and the people running the machine admit they have no plan to stop it, the rest of us deserve more than a shrug and a promise that the market will figure it out.
