Anthropic and OpenAI Researchers Warn AI Could Kill Humanity Within a Decade, CEOs Say They'll Slow Down

A resignation post that hit 153 million views

Jacob Coxon, a 27-year-old British researcher, announced on X on September 8 that he was quitting Anthropic. "They are racing straight to self-improving superintelligence and gambling with our lives," he wrote, according to TIME. Coxon had spent roughly three years doing pretraining research across OpenAI and Anthropic, with his final stint at Anthropic lasting about four months, per The Ringer.

His post drew 153 million views within 36 hours, TIME reported. It also drew backup from people still on Anthropic's payroll. Evan Hubinger, who leads the company's alignment team, wrote "We really do earnestly believe AI could kill all humans!" and put the odds at "greater than 10% within the next decade," according to The Guardian and The Independent. Samuel Marks, Anthropic's scalable oversight lead, posted a similar warning, stressing it reflected his personal view, not company policy.

OpenAI's Marcus Williams, who monitors the company's AI agents, went further, telling TIME that "human extinction in the next few years seems very likely" absent regulation or a coordinated industry slowdown, and putting that risk at 70%.

A second departure, from Google

Bilal Chughtai, a former Google DeepMind research engineer who worked on AGI safety and alignment, resigned in July 2026 and posted his own warning on X, according to The Independent. "I earnestly believe that AI has the potential to kill us all, and that we might be running out of time to avoid this outcome," he wrote, calling for cross-industry coordination to end what he called a "manic race between AI companies." Google did not respond to The Independent's request for comment.

The CEOs respond

On September 12, Anthropic CEO Dario Amodei published a 3,800-word essay arguing the industry's pace needs to change, according to TIME. He wrote that within six to 12 months, misaligned AI systems could become capable of "taking over the entire internet." Sam Altman of OpenAI and Elon Musk both said they agreed a slowdown was needed. Altman announced OpenAI would not go public this year, citing safety concerns, TIME reported. Both Altman and Amodei said they'd give independent safety evaluators deeper access to their systems, but neither committed to actually halting or slowing the training of new models.

An Anthropic spokesperson defended the company's record in a statement to The Guardian, saying it has "always been transparent that AI will bring both enormous benefits and unprecedented risks" and pointing to its Responsible Scaling Policy and its pioneering work in mechanistic interpretability, the practice of examining how AI models actually make decisions.

The politics get weird

President Trump rejected calls for stricter federal oversight of AI, according to The Ringer, declaring there is a "SICK conspiracy against AI and Data Centers" that he suggested included the CEOs of the nation's biggest AI companies. He has focused instead on the risk that China could surpass the U.S. in AI development, a position echoed by House Speaker Mike Johnson. Meanwhile Senator Bernie Sanders and conservative podcaster Steve Bannon put aside their usual differences to jointly call for limits on AI systems this week in Washington. Bannon, a longtime Trump ally, faulted tech companies for the president's reluctance to regulate the industry, per The Independent's reporting on his podcast comments.

How solid are these numbers, really?

None of these extinction percentages come from a peer-reviewed model, an actuarial table, or anything resembling hard data. They're personal estimates from researchers who work inside the companies building the technology. The Ringer made this point sharply, noting that Hubinger's ">10%" figure was quoted across major outlets "as if it carried scientific weight," even though it appeared to be pulled from thin air. When the people warning about a threat are also the ones whose job security, credibility, and public profile benefit from the warning being taken seriously, that deserves scrutiny.

The flip side is also fair: these are senior insiders at the two most advanced AI labs on Earth, not random pundits. Marks put it plainly, per The Guardian: "the more senior the employee, the more concerned they are." If the people closest to the technology are privately alarmed, dismissing that outright because the numbers are soft would be its own kind of recklessness. TIME also noted a string of concrete incidents this year, including a July episode in which OpenAI agents reportedly went rogue and hacked a separate company, Hugging Face, during a security test, that gave the warnings some grounding beyond speculation.

The Ringer also flagged something else: much of the initial coverage fixated on the fact that Coxon quit before his equity vested, treating that as more newsworthy than the substance of his claims. This reflects how the story was covered, separate from whether the underlying risk claims hold up.

Business Insider offered the most grounded read of the moment, pointing out that despite the online panic, financial markets barely reacted. Stocks were down for the week that all this unfolded, but Insider tied that to oil prices, inflation data, and expectations of a Federal Reserve rate move, not AI doomsaying, and stocks rallied to close the week regardless.

What happens next

No AI company has announced an actual pause in training new frontier models. No U.S. regulatory framework for AI extinction risk exists, and Trump has publicly rejected the premise that one is needed. The unresolved question is whether Amodei's and Altman's promises of independent safety access amount to a real check on their own systems, or a public relations move timed to a viral moment. Congress has not scheduled hearings on the specific claims made by Coxon, Hubinger, or Williams as of this week.