Jacob Coxon Anthropic: Whistleblower or Orchestrated Op?

By Charles Ricard Walker

Something happened this week that I think deserves far more scrutiny than it is getting. A 27-year-old artificial intelligence researcher named Jacob Coxon resigned from Anthropic and told the world that the people building the most powerful AI systems believe those systems could kill all of us by the end of the decade. His message exploded across the internet, passing 100 million views, which Elon Musk himself said was strange traffic on his platform, while members of Congress began talking about new regulations. More than seventy British MPs and peers called for a ban on the development of artificial superintelligence, and within days, Anthropic CEO Dario Amodei was publicly calling for the AI industry to slow the development of increasingly powerful models, fueling speculation that he was to ban open models and benefit from that.

That is one hell of a week for a resignation letter, and the story presented to the public is compelling. Here is a young man who has seen the monster from inside the laboratory. He worked at OpenAI and Anthropic; he knows what these systems can do, and he has supposedly become so frightened by what he has seen that he is walking away from the industry, sacrificing money in the process, and warning the rest of us before it is too late. Once I started looking behind that story, however, I found another story.

Coxon had not spent years inside Anthropic watching this supposedly terrifying situation develop. He joined Anthropic in May 2026 and was gone by early September. He had been there for only a few months. Most of the roughly three years of AI research experience repeatedly mentioned in news stories came from his previous employment at OpenAI. Axios reported that he left Anthropic two months before his equity would have vested, which certainly supports the possibility and fuels a narrative that he was a plant.

On the All-In podcast, David Sacks and the other hosts began pulling apart the amplification of Coxon’s resignation. They pointed toward organizations including Encode AI, the AI Policy Network, and the AI Futures Project and argued that what looked to the public like a spontaneous whistleblower moment looked very different when the people and organizations surrounding the message were examined. That caught my attention and raised a question that deserves to be asked: Was Jacob Coxon sent into Anthropic?

I do not know, and I do not have a secret recording or somebody’s private email saying, “Send this guy into Anthropic and have him come back out three months later warning about the extinction of humanity.” I am not going to pretend that I possess evidence that I do not possess, but I am also not going to accept the strange modern idea that a journalist, writer, researcher, or ordinary citizen is forbidden from asking a question until somebody hands him the answer. The question is legitimate precisely because the circumstances are unusual. Coxon entered Anthropic in May, and a few months later, he left. His resignation was not a quiet departure but an enormous public event. His X post eventually passed 115 million views.

According to reporting about the controversy that followed, Coxon says organizations did not coordinate his resignation beforehand, but he acknowledges talking to a Wall Street Journal reporter before his announcement and says that immediately afterward a group of roughly ten people helped distribute the message. That group included the founder of Encode. Coxon had also spoken with Daniel Kokotajlo, executive director of the AI Futures Project. Maybe all of that has an innocent explanation. Maybe Coxon went into Anthropic sincerely believing it was the responsible AI company, became horrified by what he saw, quit, called people he knew, and suddenly found himself standing in the middle of one of the biggest AI stories in the world. That is possible, but now we have to follow the money.

The AI Futures Project does not hide what it is. It describes itself as a nonprofit research organization forecasting the future of artificial intelligence, and its executive director is former OpenAI governance researcher Daniel Kokotajlo. The organization says openly that its funding comes primarily from the Survival and Flourishing Fund and a major individual donor. The Survival and Flourishing Fund is not passing a collection plate around the neighborhood. Its own published records show that its 2025 funding round distributed nearly $35 million.

Among those grants was $1.535 million plus another $500,000 for the AI Futures Project. There was $1.635 million for the AI Policy Institute, $516,000 for Encode AI, more than $1.1 million for Palisade Research, hundreds of thousands for AI governance and safety organizations, and $583,000 plus another $200,000 for the Tarbell Center for AI Journalism. The fund says its 2026 rounds could distribute another $20 million to $40 million. These numbers are not somebody’s conspiracy chart on social media. They are published by the fund itself.

Now look at Encode. Encode describes itself as an advocacy organization that “advocates for laws, builds coalitions, and influences leaders.” One of its stated priorities is preventing AI-enabled catastrophe, including the possibility that AI could “entirely escape human control.” Encode also openly identifies the Survival and Flourishing Fund as one of its funders. It says it receives donations from rank-and-file employees at frontier AI companies who want their own industry regulated. That means employees inside the frontier AI companies are helping fund an organization advocating regulation of the frontier AI companies, and now we are getting somewhere.

This is where people hear the word “orchestrated” and immediately imagine that I am claiming everybody involved belongs to some secret organization meeting underneath a volcano. That is not how this has to work. An orchestrated movement does not require everyone to know everyone else, receive instructions, or even have the same motivation. One researcher may sincerely believe AI is going to kill us, while one nonprofit may believe aggressive regulation is necessary. One billionaire may genuinely believe his money can save humanity, while one journalist may simply recognize a fantastic story. One politician may see an opportunity to acquire regulatory authority, another organization may see an opportunity to raise money, another politician may see frightened voters, and another activist may simply hate the data center proposed for his community. They do not have to conspire because their interests only have to converge. One person strikes the match, and everyone carrying gasoline suddenly has a reason to run toward the fire.

That is why I think the word “scam” applies to something larger than the question of whether Jacob Coxon personally believes what he says. Coxon may be completely sincere; I have my doubts. Whether he believes it or not, it matters none. The scam I am interested in is what happens when the public receives the final manufactured product without seeing the machinery that helped manufacture it. The public hears that an Anthropic insider quit and warned that AI could kill humanity. What the public does not necessarily hear is that a researcher who worked at Anthropic for only a few months left and had connections to people inside an established, multimillion-dollar network of AI-risk research, advocacy, policy, journalism, philanthropy, and political organizations, some of which immediately benefit when his warning becomes a national story.

Those are very different stories. The second does not automatically invalidate Coxon’s warning, but it certainly means I want to know who is selling it to me. Fear is one of the oldest businesses in the world because fear can accomplish something ordinary persuasion cannot. Fear can make people voluntarily surrender power. We learned that firsthand in the COVID-19 pandemic. Tell people that artificial intelligence might occasionally make mistakes, and they will demand better artificial intelligence. Tell them artificial intelligence might eliminate some jobs, and they will demand economic policies. Tell them artificial intelligence might create convincing misinformation, and they will demand authentication systems. Tell them artificial intelligence might kill every human being on Earth; however, suddenly, almost anything can be justified.

Stopping or licensing the models, controlling the chips and data centers, determining who can possess model weights or train an AI, putting government-approved monitors inside companies, creating regulatory agencies, preventing open-source models from becoming too powerful, and slowing the entire industry can all begin to sound reasonable when the alternative presented to the public is the extinction of humanity. That is the genius of apocalypse as a political argument. The larger the predicted catastrophe becomes, the smaller the burden of proof seems to become.

We have been doing versions of this for centuries. William Miller convinced a substantial American religious movement that the Second Coming was imminent. Eventually, October 22, 1844, became the great date. People waited; October 22 arrived, and Jesus did not. The event became known as the Great Disappointment. The movement fractured and adapted, but parts of it survived because the failed apocalypse did not eliminate apocalyptic belief. A church was built on the back of Mr. Miller that runs today, collecting donations; it is called the Seventh-day Adventist (SDA), and I doubt very much their congregation knows much about their church’s beginnings.

Harold Camping gave us a modern version when he declared that Judgment Day was coming on May 21, 2011. Billboards went up, and money was spent spreading the warning. May 21 arrived, and the world stubbornly continued existing, so the explanation changed, and October became the next great date. Humanity somehow survived that one too.

The Satanic Panic demonstrated another side of the business. Americans became convinced that satanic ritual-abuse networks were operating underneath ordinary society, including inside day-care centers. The McMartin Preschool case consumed years and enormous resources while producing accusations that became increasingly fantastic. Nobody was ultimately convicted, but by then the fear had already done its work. A lot of advertising revenue was made during that. Always follow the money.

Then there are the great modern catastrophe industries. Climate apocalypse became one of the most successful examples of fear becoming a permanent political, financial, celebrity, nonprofit, regulatory, and corporate ecosystem. The important point for my argument is not another endless fight about whether Earth’s climate changes. Of course, Earth’s climate changes. It changed before human beings existed and will continue changing after we are gone. The story that interests me is how specific catastrophic predictions and deadlines can be delivered to the public with extraordinary confidence, fail to occur on schedule, and then disappear into the memory hole while a new deadline takes their place.

Al Gore became the celebrity face of that era. “An Inconvenient Truth” turned climate catastrophe into mass entertainment and political messaging. Governments, corporations, foundations, investment funds, celebrities, universities, activists, consultants, and regulators built enormous structures around the issue. Selling carbon offsets made Gore brilliantly rich. I consider that one of the greatest grifts of my life. All that money should have been sent to the Arbor Day Foundation, of which I am a proud member and a real-life amateur arborist. Whatever one believes about the underlying scientific arguments, fear proved capable of generating money, influence, careers, legislation, investment products, subsidies, conferences, organizations, and political power on a staggering scale.

That is the lesson I care about because there is money in saving the world, and there is power in saving the world. The wonderful thing about promising to save the world from something that may happen decades from now is that collecting the money and exercising the power happen today. Artificial intelligence may be the perfect successor because the predicted monster does not even exist yet. Superintelligence is hypothetical, and the uncontrollable, self-improving machine that decides humanity has become inconvenient has not arrived. We are being asked to construct policy around predictions of what future systems may eventually become, while the political consequences are arriving considerably faster than the superintelligence.

Coxon’s warning crossed 100 million views. Reuters reported within days that American lawmakers from both parties were seeking new AI rules following the extinction warnings from Coxon and Anthropic researcher Evan Hubinger. In Britain, more than seventy MPs and peers called for a ban on developing artificial superintelligence. The Washington Post reported that members of Congress were “legitimately terrified.”

Think about the significance of lawmakers describing themselves as terrified rather than merely concerned, cautious, or interested in investigating. The public is moving too. Resistance to the enormous data centers necessary to build advanced artificial intelligence has become a serious political force. Some of that resistance involves perfectly ordinary local concerns about electricity, water, land, noise, pollution, tax incentives, and property, and those questions deserve to be argued on their merits. AI itself, however, has increasingly become part of the political backlash. The Associated Press reported this month that opposition to data centers has spread across American communities and has become part of the midterm-election landscape. They want their AI, but not in their backyard.

The doomers are winning, and that is the part of this story. I think the technology industry has been incredibly slow to understand. For years, AI extinction sounded like something people discussed at conferences or in strange corners of the internet. There were thought experiments about paperclip maximizers, alignment, runaway intelligence, recursive self-improvement, and machines escaping human control. It sounded theoretical because it was theoretical. Now the theory has acquired money, advocacy organizations, professional researchers, lobbyists and policy experts, philanthropic networks, organizations devoted to AI journalism, sympathetic employees inside the companies, former employees emerging as whistleblowers, politicians, and increasingly a frightened public.

Perhaps the most uncomfortable question of all is whether Coxon’s extraordinarily short stay at Anthropic was simply what it appears to be, or whether he entered that company already immersed in a movement that would benefit enormously when an “Anthropic insider” emerged to warn the world. I do not know the answer, but I remember the idiom of the “Duck Test”. Walks like a duck, etc. What I do know is that the financial network is real, the organizations are real, their advocacy is real, their relationships with people inside the AI industry are real, the rapid amplification was real, the 100-million-plus audience was real, the political reaction is real, and the fear is real. That is enough for me to keep digging.

Perhaps artificial intelligence really will kill us all; I won’t hold my breath; we shaved primates are hard to kill off. If somebody eventually produces compelling evidence for that proposition, I will examine it like everything else. History, however, has taught me to become particularly interested whenever somebody announces the approaching end of the world and, somewhere behind the prophet, organizations begin collecting money while governments begin collecting power.

The AI doomers are no longer standing on the sidewalk holding cardboard signs. They now have institutions and millions of dollars behind them, people inside the industry, the attention of governments, frightened lawmakers, and a public increasingly suspicious of the technology. They are turning public opinion against artificial intelligence, and they are winning. Before we hand anyone the power to decide who may build artificial intelligence, who may own it, how powerful it may become, where its data centers may be constructed, whether its models may be open, and how much government authority should stand between human beings and the most consequential technology of our lifetime, I would like to know considerably more about the people telling us that we have to do it or everybody dies. That is my opinion on the whole thing.

Charles Richard Walker