Some stories go viral because they genuinely hit a nerve with the public, while others seem to arrive on the scene already perfectly packaged, with the right protagonist, the right message, the media already in position, the amplifiers ready to go and, almost magically, politicians already holding the legislative answer to the problem that has supposedly just exploded into public view. The Jacob Coxon story, at least from the way it is developing, looks increasingly like the second kind rather than the first.
The narrative being presented to us is simple and extremely powerful: an Anthropic researcher leaves one of the most important companies in the AI industry because he is deeply concerned about the risks posed by artificial intelligence, breaks his silence and warns the world that we are not prepared for what is coming. It is almost the perfect story. There is the whistleblower, the giant technology company, the existential threat and the individual courage of someone willing to walk away from a prestigious job because what he has seen worries him so much that he feels compelled to speak. The problem begins when you stop looking only at Coxon and start looking at everything that happened around him, because at that point it becomes increasingly difficult to tell where spontaneous public interest ends and organised amplification begins.
Coxon effectively came from nowhere on social media and, within an extraordinarily short period of time, his message reached staggering numbers, passing one hundred million views and attracting hundreds of thousands of followers. We are not talking about Elon Musk, Sam Altman or somebody who already had a massive public profile and a well-established media machine behind them. We are talking about an account with almost no previous activity that suddenly found itself at the centre of the global debate about artificial intelligence. That kind of growth is, at the very least, extremely difficult to reconcile with what we would normally consider an organic social-media trajectory.
Taken in isolation, of course, this would prove very little. The internet creates bizarre viral phenomena all the time and it would be intellectually dishonest to argue that a large number of views is, by itself, evidence of manipulation. Algorithms produce strange outcomes, people pile onto stories unexpectedly and sometimes something simply catches fire. If the numbers were the only unusual part of the Coxon story, there would be very little more to say. The problem is that they are not.
Almost simultaneously with Coxon’s post, the Wall Street Journal publishes an article about his resignation. Then come the reactions and amplifications from people directly involved in AI policy, AI safety and regulation, with some responding within six, nine or fifteen minutes of the original publication. These are not random social-media users who happened to discover a post after it had already become viral. They are people who professionally inhabit the world of artificial-intelligence policy and organisations that have spent years arguing for stronger restrictions on the development of the most advanced AI systems. At that point the interesting question is no longer whether somebody happened to retweet something unusually quickly. The interesting question is how so many people belonging to the same political, professional and financial ecosystem happened to be perfectly positioned to amplify the same message almost the instant it appeared.
There are perfectly innocent explanations for that. They may have known about the Wall Street Journal article before publication, they may operate inside the same professional networks, somebody may have informed them in advance or they may simply monitor anything connected to Anthropic extremely closely. All of that is entirely possible. What makes the story considerably more interesting, however, is that these people and organisations are not merely aligned around the same ideas. They are also connected through a remarkably concentrated network of funding.
Among the organisations appearing in the material are AI Futures Project, Encode AI and AI Policy Institute, several of which receive direct or indirect funding from the same relatively small group of major donors and philanthropic organisations, including Survival and Flourishing Fund, Good Ventures and Coefficient Giving, formerly known as Open Philanthropy. Some of the funders mentioned also have connections to Anthropic as investors. None of this proves that somebody sat in a room and said, “Right, let’s manufacture the Coxon story.” There is no leaked email, no operational document, no instruction to Coxon and no direct evidence showing that somebody centrally coordinated the entire sequence. Claiming that as an established fact would be dishonest. What the network does show, however, is something that should be more than enough to justify serious questions: the apparently spontaneous message of a single researcher was immediately picked up and amplified by an organised, professional and well-funded ecosystem that already had a very strong interest in influencing public policy around artificial intelligence.
This is where the contours of a PSYOP begin to appear, and by PSYOP I do not necessarily mean intelligence agencies, secret rooms or operatives receiving orders from some hidden controller. In its broader and much more useful sense, a psychological or influence operation is about shaping public perception through a deliberate combination of messages, symbols, perceived authority, emotion, repetition and timing. Most importantly, it does not require the underlying event to be fake. In fact, influence campaigns are generally far more effective when they begin with something real. You take a real person, a real concern and a real statement, and then place those things inside an amplification structure capable of producing a reaction vastly greater than the original event would ever have produced on its own.
It is in precisely this sense that Coxon can be understood as the puppet. That does not mean he is necessarily lying, that somebody wrote his statements for him or that he received instructions about what to say. Coxon himself, in the interview shown at the end of the video, explicitly denies working with third parties and maintains that his concerns are entirely his own. There is no reason to pretend we have evidence proving otherwise. But a puppet does not necessarily have to understand the whole performance in order to be useful within it.
Coxon can genuinely believe every single thing he says and still become the perfect vehicle for a narrative being constructed around those statements by other people. From a media perspective, the human story of an insider who leaves one of the world’s leading AI companies because he is frightened by what he sees is vastly more powerful than another academic paper, another policy report or another document produced by an AI-safety organisation. The public is no longer being asked to think about probability models, theoretical scenarios, technical assumptions or competing interpretations of long-term risk. Instead, it is given a story that anybody can immediately understand: “I worked inside this industry, I saw where this was going and I left because I thought it was too dangerous.” That is an extraordinarily effective narrative, precisely because it turns an abstract technological and political argument into something emotional, personal and immediate.
What makes the whole sequence even more significant is what happens on the political side. According to various reconstructions, at least twenty-two politicians publicly intervened in the Coxon story, while particularly severe legislative proposals concerning artificial intelligence and so-called superintelligence were already available. That matters because we are not looking at the normal sequence in which public concern develops, politicians begin discussing the issue and legislation is eventually drafted months or years later. In this case the political window already exists, the proposed solution is already on the table and the organisations advocating for it are already in place. What is still needed is enough public pressure to make the solution appear urgent and necessary.
Then the sequence becomes extremely difficult to ignore. First comes the individual capable of putting a human face on the danger. Then comes a major newspaper giving the story credibility and reach. Almost immediately afterwards comes a network of AI-policy figures, researchers and advocacy organisations amplifying the message. The story explodes across social media, politicians react and legislative proposals that already existed suddenly find themselves being discussed inside an emotional environment that looks very different from the one that existed only a few days earlier.
If the same structure appeared in a commercial marketing campaign, nobody would have the slightest difficulty recognising what they were looking at. We would call it a coordinated launch, media strategy, influencer activation, earned-media amplification and political outreach. These are completely normal concepts when somebody is selling a product, launching a company or trying to shape the market around a commercial message. Yet when essentially the same structure appears around the formation of public opinion and around legislation capable of deciding who will be allowed to develop advanced artificial intelligence in the coming years, we are suddenly expected to believe that every element simply happened to fall into place by accident.
There is another part of the story that deserves just as much scrutiny, and that is the relationship between funding and the information ecosystem itself. When the financial connections between some of the same major philanthropic organisations, journalism fellowship programmes, media initiatives and organisations promoting the existential-risk narrative around AI are reconstructed, further links begin to appear. The Tarbell Fellowship Program is one of the examples mentioned in the material, together with funding directed towards editorial organisations and initiatives involved in communicating AI-safety concerns.
Again, the point is not to claim that a journalist who receives money through a fellowship is therefore corrupt, nor that every article published by an outlet that has received philanthropic funding should automatically be considered propaganda. That would be an absurd simplification. The issue is transparency. If the same financial ecosystem is simultaneously supporting advocacy groups, researchers, policy organisations, experts and journalism programmes that communicate with the public about exactly the same issue, then the public should be able to see those relationships clearly. Otherwise what appears to be a large number of independent voices may in reality be a collection of different nodes inside the same network.
This, in my view, is probably the most important part of the entire story, because modern influence operations do not need a single official propaganda source. In fact, that would make them considerably less effective. The far more powerful method is to create the appearance of spontaneous consensus. A researcher makes a statement, a newspaper reports it, an expert comments on it, an NGO amplifies the expert, a politician reacts, another newspaper reports on the politician’s reaction and social media then distributes the whole chain to millions of people. To the ordinary reader these appear to be separate and independent confirmations of the same reality, and the natural conclusion is that if so many different people and institutions are saying the same thing, the underlying claim must be broadly accepted and therefore probably true.
But if many of those apparently independent voices are connected through the same funding networks, the same professional circles and the same regulatory objectives, then what appears to be spontaneous consensus may actually be something very different. It may be manufactured consensus.
That leads to the question that should particularly interest anybody who actually builds technology rather than merely follows political arguments about it: who benefits from extremely heavy regulation of artificial intelligence? The most obvious answer is the public, assuming those regulations genuinely protect people from a real danger. But there is another perfectly plausible answer, and that is the companies that are already enormous, already heavily capitalised and already equipped with the infrastructure, legal departments, compliance teams, political relationships and billions of dollars necessary to absorb whatever regulatory burden governments decide to impose.
For a giant company, licensing requirements, audits, certifications and compliance obligations may simply become another cost of doing business. For a startup, an independent research laboratory or a new competitor, exactly the same requirements can become an insurmountable barrier to entry. A regulatory structure that costs a major corporation a few more lawyers and compliance officers can kill a smaller company before it has even had the opportunity to compete. Economic history contains plenty of examples of regulations introduced in the name of controlling large operators that ultimately strengthened those very operators because they were the only organisations capable of absorbing the cost and complexity of the new rules.
None of this means Anthropic organised the Coxon affair. The material analysed does not contain sufficient evidence to make that claim as a fact, and pretending otherwise would weaken rather than strengthen the argument. What it does mean is that the idea of a large technology company benefiting from regulations that dramatically increase the cost of entering its market is hardly some exotic conspiracy theory. There is an established term for precisely this phenomenon: regulatory capture. It is one of the oldest mechanisms through which powerful incumbents can turn regulation that is supposedly designed to control them into a competitive advantage.
The irony is difficult to miss. The entire operation is presented as an attempt to protect society from the uncontrolled power of the giant AI companies, yet the eventual result of the policies being proposed could be a system in which only those same companies have enough money, infrastructure and political access to continue developing advanced artificial intelligence. The regulation supposedly intended to restrain them could end up building the walls that protect them from future competitors.
That is also why Coxon himself is, to me, considerably less interesting than the people and organisations surrounding him. Whether he is sincere is almost beside the point, because he may very well be completely sincere. Whether somebody suggested what he should write is also secondary, because it is entirely possible that nobody did. The more important questions concern what happened after he spoke: who transformed his statements into one of the dominant stories of the moment, who was ready to amplify them, who finances those organisations, what relationships exist between those funders and the AI industry and, most importantly, what political decisions those organisations are trying to obtain.
When a virtually unknown individual publishes a message and, within hours, major media organisations, advocacy groups, policy experts, financial backers and dozens of politicians appear almost perfectly aligned around the same narrative, eventually the explanation that “it simply went viral” begins to require more faith than scepticism. Coxon may indeed be nothing more than a genuinely concerned researcher who decided to speak publicly about something that frightened him. That possibility should not be dismissed.
But what was built around him has all the hallmarks of something considerably more sophisticated.
And when fear becomes the instrument used to persuade the public to hand governments, regulatory agencies and giant corporations the power to decide who will be allowed to develop artificial intelligence and who will effectively be prevented from doing so, the first question we should be asking is not how frightened we are supposed to be of AI. The first question should be who is trying to make us afraid, who is paying to make sure that message reaches everywhere, and, above all, what they hope to obtain once enough people are frightened enough to accept it.