Follow the Money with the AI Debate
There is something strange about the current debate over artificial intelligence. Some of the people building the world's most powerful AI systems are warning us that the technology they’re developing, that will make them among the wealthiest humans on the planet, could also soon destroy humanity. OpenAI's Sam Altman and Anthropic ‘s Dario Amodei have been among the most prominent voices warning about potentially catastrophic AI risks. AI engineer Jacob Coxon, who spent three years at OpenAI and Anthropic, just resigned from Anthropic because he believes the risks are increasing faster than the industry’s ability or willingness to control them. Coxon accused both companies of racing toward self-improving superintelligence and “gambling with our lives”.
Anthropic alignment lead Evan Hubinger publicly agreed with Coxon’s warning and put his own estimate of AI killing all humans within the next decade at greater than 10 percent. The fact that Coxon was sufficiently concerned to walk away from Anthropic, before his Anthropic equity vested, makes his warning a bit harder to dismiss. Then again he is already wealthy, and maybe he thought this would significantly increase his status within the AI community may present new opportunities.
Right on cue some politicians and their politically aligned media chimed in and demanded legislation to control AI. Within days after Coxon’s resignation, Barack Obama weighed in, warning of “potential catastrophe” and calling for federal laws and regulations governing AI safety. Senator Elizabeth Warren went further, calling for a pause in advanced AI development and arguing that its future should be decided by elected government rather than “a handful of CEOs and billionaire corporations.” Bernie Sanders and his Democrat Socialist Party go even further, as they want the federal government to own controlling interests in all sorts of nationally strategic companies including AI.
The media response was almost immediate. The Washington Post announced that the “Political world erupts as AI researchers warn of ‘extinction’ threat,” while The Guardian led with “Anthropic researchers say AI could cause human extinction by 2030.” CBS warned that humans are “close to being outsmarted” by superintelligence, Fortune declared “The AI apocalypse warning that finally went mainstream,” Within days, the Associated Press carried the story nationally under the headline “New warnings about the risks of AI to humanity revive a long-running debate,” while The Guardian returned with another piece asking, “Could AI really wipe out humanity?” In less than a week, Coxon’s resignation helped turn a speculative extinction scenario from an argument inside the AI-safety community into a mainstream political and media story that promises to be front and center in the midterm elections.
Not everyone is buying the AI-doomsday narrative. President Trump has gone furthest in the opposite direction, dismissing the push for new AI regulation as a “hoax” and arguing that slowing American development would hand an advantage to China. Mark Zuckerberg reached a similar conclusion from a less political direction. The Meta CEO argues that competition, liability and companies’ own incentives provide strong reasons to develop AI safely without an industry-wide slowdown and has warned that concentrating control of superintelligence in a few governments or institutions may itself present the greater danger.
In a recent Wall Street Journal essay, Zuckerberg said it plainly:
“The important question is not simply whether superintelligence arrives, but whether it becomes centralized and restricted or is broadly available as a tool for individual empowerment.”
The divide is becoming stark. Altman, Amodei and others are warning that the race may need federal government restraints, while Trump, Zuckerberg, and Nvidia’s Jensen Huang are questioning whether the proposed cure, slower development, and more regulation, could cause missed opportunities and cede AI leadership to our main adversary, China.
Maybe the AI-Doomers are right but there's another possibility that deserves more scrutiny. The companies leading the AI race may be helping create the political conditions for a regulatory regime that they are uniquely equipped to survive. That doesn't require believing that Sam Altman and Dario Amodei secretly think AI is perfectly safe. It doesn't require believing that every scientist warning about AI is dishonest. And it certainly doesn't require believing that artificial intelligence presents no serious risks. It requires asking a much older question.
Cui bono? Who benefits? a.k.a. Follow the money.
That question is where any investigation should begin.
What is a Psyop?
Psyop is short for “psychological operations.” It is an organized effort to influence what a target audience believes, feels, decides, or does. It’s a form of subversive coercion, but unlike physical violence, psyops target the mind and decision-making process. It contains truthful information, misinformation, cherry picking facts, emotional appeals, propaganda, rumors, or some combination of these. Historically, militaries and governments have used psychological operations to influence enemy troops, civilian populations, allies, or other groups.
Psyop has become one of those words that gets thrown around until it means almost nothing. An advertisement somebody doesn't like is a psyop. A political slogan is a psyop. A TikTok trend is a psyop. Real psychological operations are more complex. The historical record shows that successful influence operations rarely depend upon convincing a population of something entirely imaginary. Often the more effective technique is to begin with something real and shape how people understand it.
The British-lead and CIA-backed operation against Iranian Prime Minister Mohammad Mossadegh in 1953 is a good example because much of the record has been declassified. The operation included propaganda, political pressure, press manipulation and what the CIA itself described as a "war of nerves." During the Cold War, American intelligence agencies also used radio, publications, civic organizations, intellectuals and other intermediaries to influence foreign populations.
In addition to the CIA Iran psyop, the 1975 Church Committee eventually exposed domestic abuses as well, including COINTELPRO operations intended to disrupt and discredit organizations and individuals the FBI considered threatening. None of this is controversial conspiracy theories. It’s history. And history tells us something important about how influence works and how the intelligence community deep state can and does conduct psyops both foreign and domestically.
The public doesn't necessarily have to be told an outright lie. A genuine danger can be emphasized while mitigating facts are ignored. An uncertain proposition can gradually be reported as a likely one. The same message can appear through multiple channels, giving people the impression that many independent sources have arrived at the same conclusion. Most importantly, the purpose isn't simply changing belief but changing behavior.
The People Selling the Product Are Warning Us About the Product
Imagine Pfizer announcing that a new drug it was developing might save millions of lives but might also accidentally kill 10% of the world's population. Now imagine their proposed solution was a federal regulatory system determining which pharmaceutical companies were “aligned” enough to develop similar drugs. Most journalists would immediately recognize the potential conflict of interest. Yet something similar is happening in AI. The public is being asked to evaluate technical predictions about systems almost nobody outside a small group of specialists understands. Politicians don't understand them either. Neither do journalists. So naturally they turn to “the experts.” This groupthink can produce rhetoric that seems weird and disturbing to the general public.
And it’s no secret that some of the high-profile experts belong to a community of exotic belief and practice that informs their group think. In the early days of AI, this community rallied around a movement called “Effective Altruism” (EA). These were a group highly intelligent nerdy but persuasive young men, that included Amodei, FTX crypto felon Sam Bankman-Fried, 23-year-old Leopold Aschenbrenner, the founder of a $78 AI-focused hedge fund that recently blew up, and a gaggle of self-styled philosophers, investors and engineers. Through his FTX Foundation, Bankman-Fried was a $500 million early investor in Anthropic. His stake was sold to partially settle the fraud claims he was convicted for.
Unfortunately, many of the world's leading experts work for the companies building the systems. This gives frontier AI companies an extraordinary position in the political debate. They are simultaneously developers of the technology, experts on its capabilities, authorities on its potential dangers, and participants in designing the regulatory solution.
But there's another question beyond the regulation’s intent. What does it actually do to competition? OpenAI, Anthropic, Google and Meta have enormous amounts of capital. They have lawyers, lobbyists, cybersecurity departments, policy staffs, compliance teams and relationships with governments around the world. The next two kids building an AI company in a garage don't. A $10 million regulatory burden doesn't affect those organizations equally. To a company spending hundreds of billions of dollars developing frontier models, it may amount to another line item. To a startup, it can mean the company never gets started.
This is hardly a new thing. Industries have been using regulation as a competitive moat for generations. Big banks can absorb banking regulation that destroys small banks. Prior to the 2008 Financial Crisis we had about US 8,500 banks and today we have about 4,300, a 50% decrease. Big pharma companies can navigate approval processes that smaller competitors cannot afford. Big corporations routinely discover that once government regulation becomes inevitable, helping write the regulations can be much more profitable than opposing them. The right question isn't whether OpenAI and Anthropic favor "AI safety." Of course they do. The question is what the regulatory system built in the name of AI safety will do to the structure of the industry ten years from now.
Then There Is the Press
A psychological influence campaign needs amplification. It doesn't necessarily need willing accomplices. It doesn’t need an editor at CNN to get a telephone call from OpenAI headquarters every morning explaining what to report. Though it’s worth pointing out we now know the Biden administration was doing just that in the 2020 election cycle stationing FBI agents and political operatives in the offices of Facebook, Twitter and other social media companies.
But AI is more complicated. A reporter assigned to cover it cannot independently evaluate whether a new model represents a 1% or 20% extinction risk. When prominent scientists and CEOs say their systems may become catastrophically dangerous, that's a story. The Reuters Institute at Oxford studied AI coverage in six mainstream British news organizations and found that almost 60% of stories focused on products, announcements and initiatives. About a third of identifiable sources came from industry, about six times the proportion from government and nearly twice that from academia. Its broader research described AI journalism as frequently industry-led and willing to repeat speculative claims about future capabilities. Think about the implications.
The AI industry can generate both hype and fear. A company announces an astonishing new capability. Newspapers report that AI is advancing faster than expected. A company executive warns that continued advancement could become dangerous. Newspapers report that experts fear catastrophe. Politicians demand hearings. The hearings become news. International organizations hold conferences. Those conferences produce reports. The reports are cited by the next round of journalists. Eventually, a prediction originating within a relatively small community begins to look like an independently established consensus. Nobody has to coordinate the whole thing. Each participant only has to behave according to his own incentives.
That is what makes modern information cascades so powerful.
COVID Should Have Taught Us Something About Certainty
The COVID experience is relevant here, although not for the reason people on either side of that argument usually assume. COVID was real. It killed millions of people. Nothing about questioning government or media behavior during the pandemic requires pretending otherwise. The interesting issue is what happened to uncertainty.
During the early pandemic, the possibility that SARS-CoV-2 might have originated with laboratory research in Wuhan China became extraordinarily contentious. Public discussion mostly treated natural zoonotic emergence as the scientifically likely explanation while describing a laboratory-related origin as conspiratorial. We now know that the private discussion among scientists and government officials was considerably less certain. Recently released government material has generated new proof about how the laboratory-origin possibility was improperly handled by Dr. Robert Fauci and his staff inside the U.S. government.
Meanwhile, the scientific question remains unresolved. A group of WHO origins researchers writing in Nature in 2026 argued that the balance of published scientific evidence continues to favor zoonotic emergence while acknowledging that important information needed to establish the origin conclusively remains unavailable. It’s worth pointing out that China carries immense weight at the WHO and they have refused access to critical records from the Wuhan Lab that could help answer the question. The lesson isn't that we now know with certainty that COVID escaped from a laboratory. The lesson is that the range of possibilities being privately considered by knowledgeable people was wider than most of the public conversation suggested.
The pandemic demonstrated something else also. Populations in western democracies accepted government interventions that would have been politically unimaginable months earlier. Lockdowns, travel restrictions, contact tracing, vaccine documentation, business closures and emergency executive orders. That doesn't prove COVID was created to test population control, but it certainly demonstrated something governments have understood for centuries. That an emergency changes what people will tolerate from government.
In the immortal words of Obama campaign manager and Democrat party operative Rahm Emmanuel. “Never let a crisis go to waste.” We should remember that when the emergency isn’t a virus capable of killing a small percentage of infected people, but a machine supposedly capable of killing everyone.
The Russiagate Precedent
The Russia election interference controversy following the 2016 election offers another lesson about the difference between an underlying fact and the narrative constructed around it. As it has done for many election cycles, Russia attempted to interfere in the 2016 American election. It was mostly penny ante social media influence. The Democrat-led Senate Intelligence Committee documented Russian influence efforts and special prosecutor Robert Mueller investigated contacts between Russians and individuals associated with the Trump campaign. But he could not prove that anyone in Trump campaign abetted or knew about the Russian interference. Those distinctions were blurred during years of extraordinary media attention. Subsequent investigations and declassifications have exposed how intelligence was manufactured by the Hillary Clinton campaign in coordination with the Obama administration intelligence community. A classic psyop.
Russiagate illustrates the larger phenomenon. The underlying event doesn't have to be fake for the resulting narrative to become misleading. It takes something that happened, select the most useful interpretation, repeat it relentlessly and allow people to infer much more than the evidence establishes.
Follow the Money Through the NGOs
There is another component to modern influence that deserves much more attention: the enormous ecosystem of foundations, NGOs, think tanks, activist organizations and ostensibly independent policy institutions. CCP-connected billionaire and China resident Neville Roy Singham is a striking example. A New York Times investigation traced hundreds of millions of dollars through organizations connected to Singham and reported extensive relationships between that network and Chinese state-media and propaganda interests. Organizations within the network shared personnel and infrastructure while promoting positions favorable to Beijing and the Democrat party.
Foreign funding doesn't automatically turn an American organization into a foreign agent. FARA has specific requirements concerning relationships with foreign actors. But step away from the legal question and consider the information architecture. An American watching an American activist representing an American nonprofit reasonably assumes he is hearing an American political organization expressing its independently developed views.
What if the organization's financing ultimately originates abroad?
What if several apparently independent organizations share funding sources?
What if they share personnel?
What if they rely upon one another's research?
Suddenly what appears to be five independent signals may actually be one signal passing through five speakers.
George Soros and his Open Society Foundations is another example. There is no need to speculate about secret foreign-government direction. Open Society openly spends enormous sums supporting advocacy, litigation, research and organizations seeking policy change. Its U.S. expenditures were $242 million in 2024 alone. Rather than arguing about whether those activities should be called "left," "progressive" or something else, look at the grants. Look at what the recipient organizations advocate. Look at the litigation they pursue. Look at the legislation they support.
Then ask whether the supposedly independent institutions shaping public debate are as independent from one another financially as they appear to be editorially. The same analysis should apply to conservative money. If Koch-funded organizations are advocating deregulation, readers deserve to know that too. This isn't about which billionaire is virtuous. It's about understanding who built the megaphone you're listening to.
The Global Governance Question
The World Economic Forum is frequently described by its critics in conspiratorial terms. Its basic philosophy is public. The WEF explicitly promotes public-private cooperation among governments, corporations, academics, NGOs and international institutions. It believes major global problems require coordinated solutions crossing national and institutional boundaries. You can agree with that philosophy. But you should understand what it means for democratic government.
In our American model, citizens elect representatives, representatives enact laws and governments execute them. The chain of accountability is imperfect, but it’s visible. Global governance is different. Experts, multinational companies, NGOs, international organizations and government representatives develop standards and policy consensus that can find their way into national policy. Sometimes that's perfectly sensible. But the more problems we define as inherently global, the more political authority naturally migrates toward institutions operating above sovereign nations. Look no further than European Union to see this dynamic in action.
And AI is perfectly suited to that argument. If uncontrolled artificial intelligence could genuinely destroy humanity, then maybe no individual country can be trusted to regulate it independently. One reckless nation could endanger everyone. China would be a prime candidate just as it unleashed Covid on the world either accidentally, on purpose, or some combination. Once you accept that premise, international governance stops sounding optional. It starts sounding inevitable. That's why the AI extinction narrative has consequences far beyond technology policy. It potentially provides intellectual justification for shifting decisions about one of humanity's most important technologies away from markets, individuals and national legislatures and toward an international network of regulators, corporations and technical experts.
Anyone who believes in national sovereignty ought to understand that trade before making it.
Where do I come down on this?
I had CNBC on TV in the background of my Budapest hotel as I was writing this blog. I stopped my writing and began to listen to a conversation with Chris Hughes, a Facebook cofounder. He advocates FNRA-style regulation of AI. In answering Beckie Quick’s question about the exemption social media receives from section 238 FCC regulation social media companies negotiated years ago, and how it has allowed social media growth in potentially toxic ways where they should have had the same liability as traditional TV, radio and print media.
Hughes answered that I’m sure we wish we had oversight of the kind of systemic financial risk that drove the 2008 financial crisis. Then he asked,
“Do you want to wait until there is an event, like AI shutting down our electrical grid and the subsequent catastrophe, and then trying to settle the immense economic and social costs, in court with liability lawsuit?”
I think that was a great answer by Chris even given my bias for less regulation. Almost all Americans can agree that some government is essential, but the trick is getting the right constitutional scope and balance.
With great irony 15 minutes later, there was another CNBC conversation with Jensen Huang, Nvidia’s founder. I’m literally watching the exact argument I’m having in my own brain. Huang says there is an eminently effective technical solution to this problem, and Nvidia is building it. He describes a containment system that first revokes all access to anything to AI and then add back specific use cases of access as determined by humans.
The tech industry built the web browser to be a gatekeeper for interacting with internet. It can shut down all access and grant it back as needed. Kind of a whitelist versus blacklist system. Blacklist security systems starts with granting all access and then revoking access of certain types as specified by human users. The whitelist system says there is NO entry to the club unless you are on the list. Huang is creating a hardened whitelist sandbox with which ALL AI operation resides. Like our nuclear codes, there is human chain of custody gauntlet that must be crossed before someone can push the button.
This felt like a brilliant answer that epitomizes the Occam’s razor postulate that the simplest answer is usually the right one. We have always engineered solutions to digital challenges as they have arisen. As I describe in my book Locally Grown: The Art of Sustainable Government, we did this when we hit traffic limitations with the first generation of mainframe computers. To resolve it, we decentralized the compute resources over many connected network servers. The internet is the result of that innovation. There is no AI without the internet. If we regulated the internet in the earliest stages, everything would look different today. Not just the digital world, but the entire economy. America would look like Europe, a continent with bloated unsustainable governments that cannot defend themselves.
Follow the Meta Example
How about this recent statement by Meta CEO Mark Zuckerberg said?
“Meta delayed shipping its new AI agent Muse for several months to focus on safety and security. We just did it as part of our day-to-day work because it was clearly the right thing for people and for us.”
Meta learned an important lesson in product liability when it settled lawsuits for $18 billion over teen social media addiction in August. No more moving fast and breaking things. OpenAI and Anthropic are learning fast too but they need to create safe products rather than calling for the nanny state to clean up their messes. Here is Palantir CEO Alex Karp on CNBC on AI companies:
“They’re saying the only way to deal with this kind of liability is to go to the government and say, ‘Nationalize us, please.’ Let’s instead enforce our current laws. For one offense, 18 U.S. 1030 calls for a fine or up to 10 years imprisonment. I’m not encouraging this, but a few predawn headquarter raids and you can be darn sure that AI agents won’t leave their sandboxes again. Safety first.”
Circumstantial Evidence Counts
I can hear the skeptics now. None of this proves a psyop. I have seen no public documents where Sam Altman, Dario Amodei, George Soros, Klaus Schwab and government officials sit around a conference table plotting how to frighten humanity into accepting centralized control. That's an absurd evidentiary standard. Circumstantial evidence is evidence. Courts allow juries to consider it every day because intent is rarely established through confession. It is inferred from behavior, incentives, communications, timing and outcomes.
“Cui bono” isn't a verdict. It's an investigative tool. So use it.
Who benefits from AI regulation?
Who is funding the organizations advocating it?
Who supplies the experts quoted by journalists?
Who funds those experts?
Who participates in government advisory panels?
Who writes proposed regulatory standards?
Who can afford to comply with them?
Who can’t?
Which powers will governments acquire?
Which powers will international organizations acquire?
Will those powers disappear if the predicted catastrophe doesn't occur?
And maybe most importantly, what did the people making the most frightening predictions say privately before they said it publicly?
These questions won't necessarily lead to what either the supporters or the critics of AI regulation expect. Maybe the evidence will show that the warnings are sincere, and the proposed regulations are carefully constructed to preserve competition. Fine. But that conclusion should emerge after investigation, not be granted in advance because the institutions involved possess impressive credentials. And I can guarantee this. If the Democrat party regains power they, and some Republicans, will institute exactly the kind of regulation we are warning about, and they will see to it that AI is “in alignment” with their political interests.
Could Part of the Solution be Locally Grown?
There’s another possible future for AI that gets less attention. AI doesn't necessarily have to become a handful of gigantic models running inside billion-dollar data centers controlled by the world's largest corporations and supervised by national or international regulators. Models are becoming smaller and more efficient. They can run locally. They can be open weight. Businesses can customize them. Universities can study them. Individuals can run them without asking permission from a corporate API provider. Entrepreneurs can build highly useful and productive applications without sending every query through servers controlled by three companies.
I plan on installing an open-source AI Large Language Model (LLM) on my private server, maybe Facebook Llama, and training it on my digital footprint over the last 40 years. This includes emails, texts, photos, my book manuscript, my blogs and podcasts, and my music. All this training will occur on my local server NOT connected to the internet so there is no possibility the AI agent escapes into the wild. What is the purpose of this exercise you ask? To create a pretty accurate chat bot that is me. So many years from now my ancestors exploring our digital genealogy can click a button and talk to my avatar from the grave, about what it was like in 2026.
I can hear your giggles. But having just spent a month in Europe walking through museums, palaces and churches, I would love to have been able to hear from Michelangelo what he was thinking when he painted the Sistine chapel. Not that I am in any way a talent like Michelangelo mind you, but he left an eternal footprint and I want to too. Vanity? Maybe. But as Banksy, the England-based graffiti artist and political activist famously said,
“You die twice. Once when you stop breathing, and a bit later when someone says your name for the last time.”
But I digress. This is just one of my personal use case and tens of thousands of companies around the world are using their own local versions of AI. Decentralized technologies are harder to police. They're also harder to monopolize and that may ultimately be the most important divide in the AI debate. Not AI versus no AI. Not safe AI versus dangerous AI. But centralized intelligence versus distributed “open weights” intelligence. One future places increasingly powerful artificial intelligence behind the walls of a few giant corporations operating in partnership with governments and international regulators. The other distributes increasingly capable intelligence across companies, communities and individuals. Decentralized will likely be safer while centralized will certainly be easier to control. Consider me mostly in the “Open Weights” camp. And that is precisely why we should be very careful about accepting the premise that is being sold.
Maybe Dario Amodei and Sam Altman and Elon Musk are right. Maybe artificial intelligence really could become an existential threat. But history suggests another question should always accompany predictions of catastrophe, especially when the proposed solution concentrates power in the hands of the people making the prediction.
