The day after an Anthropic researcher resigned with a public warning that AI companies are "gambling with our lives," a cascade of current employees at the company backed his claims and posted their own dire assessments. Elon Musk and other prominent voices on X responded by calling the coordinated warnings a manufactured operation designed to push regulatory agendas.
Jacob Coxon's resignation announcement on September 9 triggered what amounted to an internal revolt at Anthropic, the AI safety-focused startup behind the Claude chatbot. Within hours, employees across the company's safety and alignment teams posted publicly agreeing with his assessment that the race toward more powerful AI models has outpaced the ability to control them. The speed and breadth of the response suggest the concerns are not isolated to one departing researcher but represent a broader sentiment inside one of the industry's most prominent labs.
The Anthropic Employees Who Spoke Up
Anna Wang, who works on Artificial General Intelligence Safety at Anthropic and previously held a position at Google DeepMind, wrote that many people at the company want to slow development to develop a plan for mitigating risks. "There is not yet a viable scientific plan to solve risks from recursively self-improving AI," she posted on X.
Samuel Marks, who works on safety research at Anthropic, offered a broader observation about the company's internal culture. "AI developers believe their technology could cause human extinction, or similarly bad outcomes," he wrote. "This could happen in the next few years. In general, the more senior the employee, the more concerned they are."
Drake Thomas, another Anthropic employee, said he respected Coxon's decision to stop building models he believes could pose planet-scale risks. "Things are moving way too fast, we don't have anywhere near the degree of assurance we'll want for ASI," Thomas wrote, using the abbreviation for artificial superintelligence.
Evan Hubinger, who leads Anthropic's alignment stress-testing team, provided the most specific assessment. "We really do earnestly believe AI could kill all humans!" he wrote. "I personally think it is greater than 10% within the next decade." He added that Anthropic is trying its best but does not yet have a plan to solve alignment for superintelligence and is not clearly on track to develop one.
The pattern across these statements is notable. None of these researchers described their concerns as hypothetical or distant. They spoke in concrete terms about risks they believe are real, imminent, and insufficiently addressed by the companies building the technology. Marks's observation that seniority correlates with concern suggests the people with the deepest technical understanding of these systems are the most alarmed by what they see.
The Musk Response
Elon Musk responded to Coxon's original post with a brief assessment: "Seems like a setup." When Coxon replied with a selfie and a note that his beliefs are genuine, Musk escalated. "I think the groundwork for this psyop has been prepared for a long time," he posted. "This was just the match that lit the fire."
Musk was responding to a post by Parker Thayer, a researcher at the Capital Research Center, a conservative think tank. Thayer had proposed, with little supporting evidence, that Coxon's post represented "the beginning of a VERY sophisticated and well-funded PR operation to get support for Democrats to regulate AI into oblivion." Bill Ackman, the billionaire CEO of Pershing Square, quoted Thayer's post with a single word: "Interesting."
The psyop framing requires a specific chain of assumptions. It presumes that Anthropic employees, some of whom have spent years working on safety research inside the company, are coordinating their public statements as part of a political operation rather than expressing genuine technical concerns. It assumes that the researchers' detailed assessments of alignment challenges, recursive self-improvement risks, and the gap between capability and control are cover stories rather than descriptions of problems they actually work on daily.
The alternative explanation, that researchers at an AI safety company genuinely believe the technology they are building poses existential risks and are alarmed by the pace of development, does not require any conspiracy. It requires only that they are telling the truth about what they see in their work.
The Anthropic Response
Anthropic's official statement defended the company's approach without directly addressing the employees' concerns. "We have always been transparent that AI will bring both enormous benefits and unprecedented risks," a spokesperson told the Guardian. "To address these risks, we continue to build models with some of the strongest safeguards in the industry."
The same day Anthropic's employees were posting about extinction risks, the company released a report documenting how it dismantled an operation that had been attempting to build a biological weapon using its AI models. The report demonstrates both that the risks are real enough to require active intervention and that Anthropic has mechanisms in place to detect and prevent certain categories of misuse.
But the researchers' concerns go beyond misuse by bad actors. They are worried about the models themselves becoming more capable than the safeguards designed to control them, about recursive self-improvement creating systems whose behavior cannot be predicted or constrained, and about the competitive pressures that make it difficult for any single company to slow down without ceding ground to rivals.
The Skeptics
Not everyone in the AI field shares Coxon's assessment of extinction risk. Gary Marcus, a scientist and prominent AI critic, said the immediate dangers are already here, not hypothetical future scenarios. He called for a boycott of AI not because it might cause extinction but because it is already causing harm through AI-generated pathogens, disinformation-fueled conflicts, and attacks on critical infrastructure.
"Nothing I have seen gives any indication that any of that is under control," Marcus said. His framing suggests a different kind of urgency than the one Coxon describes, focused on present harms rather than future catastrophes, but arrives at a similar conclusion: the current trajectory is not working.
The gap between these perspectives, extinction risk versus ongoing harm, matters for policy responses. If the danger is future superintelligence, the solution might involve slowing capability research and investing in alignment. If the danger is present misuse and uncontrolled deployment, the solution might involve stronger regulation of existing systems and accountability for harm already occurring.
What Happens Next
The Anthropic employee statements represent something unusual in the AI industry: public, coordinated, technically specific warnings from people inside the companies building the most capable systems. These are not outside critics or academics speculating from a distance. They are researchers who spend their days working on the exact problems they say remain unsolved.
The psyop framing from Musk and others serves to discredit these warnings without engaging with their substance. If the researchers are part of a political operation, their technical assessments can be dismissed. If they are telling the truth about what they see in their work, the implications are much harder to ignore.
Anthropic has positioned itself as the safety-conscious alternative to OpenAI and other frontier labs. The fact that its own safety researchers are publicly stating that the company does not have a plan to address the risks it is creating represents a credibility challenge that a corporate statement about "strongest safeguards" cannot easily resolve.
The next test will be whether these warnings translate into any change in how these companies operate, or whether they become another cycle of alarming statements followed by business as usual. Coxon's prediction that things could be out of control by the end of next year sets a concrete timeline. The researchers who backed him have put their professional reputations on the line. The question is whether anyone with the power to change course is listening.