A former Anthropic researcher resigned and accused the company and OpenAI of recklessly chasing superintelligence, and a senior colleague still at the firm agreed, saying he believes there is a greater than 10 percent chance AI kills every human being within the decade.

Jacob Coxon, a researcher with three years of pre-training experience at Anthropic and OpenAI, quit his position at Anthropic on Tuesday and posted a warning on X that has since been viewed more than 70 million times. In the post, Coxon said the people building AI "earnestly believe that it could kill us all by the end of the decade" and accused both companies of "gambling with our lives," CNBC reported.

Hours later, one of Anthropic's own senior leaders confirmed the substance of Coxon's fears, publicly, on the same platform, under his own name.

Anthropic's alignment lead says the company has no plan for superintelligence

Evan Hubinger, Anthropic's alignment science lead, posted on X late Tuesday in direct response to Coxon:

"Jacob is correct here, we really do earnestly believe AI could kill all humans! I personally think it is >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."

Read that again. The man Anthropic pays to keep its AI systems aligned with human values says the company does not have a plan to solve the problem, and is not on track to find one. He still works there. He put a number on the risk: greater than one in ten.

Coxon's post was blunt about what he believes is at stake. He wrote that neither Anthropic nor OpenAI is "acting responsibly" and that both companies are "racing straight to self-improving superintelligence." Self-improving superintelligence refers to AI systems capable of upgrading their own capabilities without human direction, a threshold many researchers treat as the point of no return.

He warned the public directly:

"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."

Samuel Marks, Anthropic's scalable-oversight lead, added his own confirmation. The New York Post reported that Marks noted the more senior the employee at these AI labs, the more concerned they tend to be about extinction risk. That detail alone should give Americans pause: the people closest to the technology are the most alarmed by it.

OpenAI's own chief scientist called for voluntary slowdowns days earlier

Coxon's resignation did not happen in a vacuum. Days before he quit, OpenAI's chief scientist Jakub Pachocki published a blog post on OpenAI's website titled "An Alien Mind" that carried a remarkably similar warning. Pachocki wrote that no AI company has "solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer."

Pachocki went further, calling for the industry to pump the brakes voluntarily:

"I expect and hope for voluntary slowdowns to become commonplace until shared safety bars are established. And I believe that international coordination on future AI development needs to become a top priority for governments around the world."

So OpenAI's top scientist says the company cannot safely keep scaling at full speed, and Anthropic's alignment lead says his team has no plan to solve the core safety problem. Both men remain employed at these firms. Both firms are heading toward potentially historic initial public offerings. The financial incentives to keep building, and keep quiet, are enormous. The fact that insiders are speaking up anyway tells you something about how serious they believe the danger is.

The pattern of insider alarm is not new. In 2023, prominent AI researchers and executives, including OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei, signed a statement declaring that "mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war." That was three years ago. The companies kept building.

In a political environment where public officials face consequences for far lesser failures of stewardship, the AI industry's approach stands out: sign the warning, pocket the investment, and accelerate anyway.

Congress scrambles, but no consensus exists on how to act

The resignations and public warnings have drawn attention on Capitol Hill, though lawmakers remain far from agreement on what to do. Rep. Lori Trahan, a Massachusetts Democrat, posted on X the day after Coxon's resignation:

"Safety researchers are resigning, powerful AI models are breaking out of their labs, and companies are racing ahead anyway. It's past time for Congress to get off the sidelines and do its job."

Trahan and Rep. Jay Obernolte, a California Republican, introduced the FRONTIER Act in July. That bill aims to establish a framework for governing the deployment of advanced AI models. The Washington Examiner noted that Coxon's departure added a high-profile resignation to the broader conversation about AI safety, the kind of event that tends to move legislative timelines, at least temporarily.

Sen. Bernie Sanders and Rep. Greg Casar introduced a more aggressive measure earlier this month: the Ban Artificial Superintelligence Act, which would temporarily pause advanced AI development until the federal government establishes safety rules. Both bills have received mixed receptions.

The bipartisan nature of the FRONTIER Act, one Republican, one Democrat, suggests the issue does not break neatly along party lines. But the Sanders-Casar bill, with its outright development ban, reflects the kind of heavy-handed federal intervention that rarely produces good results and often stifles the very innovation it claims to protect. The question is whether Congress can find a middle ground that holds companies accountable without handing bureaucrats a veto over American technological leadership.

Meanwhile, roughly 1,400 AI researchers, Hubinger among them, signed an open letter in July urging the U.S. government to develop tools to "deliberately pace the frontier of automated AI development." That letter, hosted at pacingthefrontier.com, represents a significant bloc of technical expertise pressing for government involvement.

In an era when diplomatic decisions carry enormous stakes for American interests abroad, the AI safety question carries a similar weight: get the policy wrong, and the consequences could extend well beyond Silicon Valley.

Treasury Secretary Bessent puts AI companies on notice

The warnings are not coming only from researchers and lawmakers. Treasury Secretary Scott Bessent weighed in earlier this month, following G20 meetings with finance ministers and central bankers in Asheville, North Carolina. Bessent said AI companies have done a "horrendous job of explaining themselves to the American people."

He did not stop there:

"They're going to have to take some of the blame, and they are going to have to convince the American people that all the benefits will not accrue to a small group."

Bessent's comments reflect a growing concern that cuts across ideological lines: the AI industry is building something it openly admits it cannot fully control, and the public has almost no say in the matter. The National Republican Senatorial Committee said last month that AI data centers have become a "sleeper issue" for the entire midterm election cycle, a sign that voters are paying attention even if Washington has been slow to act.

That political reality matters. When constituents lose confidence in the people making decisions that affect their safety, they find ways to make their displeasure known at the ballot box.

Silicon Valley's long-running debate just went public in a way it cannot take back

The AI safety debate has simmered inside Silicon Valley for years. What changed this week is the sheer directness of the admissions, and who made them. This was not an outside critic or a disgruntled former intern. Coxon spent three years doing pre-training work at the two most prominent AI labs on the planet. Hubinger holds one of the most sensitive safety roles at Anthropic. Pachocki is OpenAI's chief scientist.

These are the people who understand the technology best, and they are telling the public, in plain language, on a public platform, that the companies they work for have not solved the fundamental safety problem and are pressing ahead regardless.

Coxon's post drew more than 70 million views. That number alone suggests the public appetite for honest answers from the AI industry far exceeds what the companies have been willing to provide. Bessent was right: they have done a horrendous job explaining themselves. But the problem may be worse than poor communication. The problem may be that the honest explanation, "we think there's a meaningful chance this kills everyone, and we're building it anyway", is not one any PR department can spin.

The pattern is familiar to anyone who has watched powerful institutions prioritize their own interests over public accountability. The difference is the stakes. A botched policy or a cynical political maneuver costs money, trust, or elections. An uncontrolled superintelligent system, by the admission of the people building it, could cost everything.

Americans deserve better than a system where the people building the most powerful technology in human history shrug, assign it a one-in-ten chance of ending civilization, and keep cashing their stock options. If the experts are right about the risk, accountability cannot wait for the IPO.