Conceptual illustration of a wide open gate leading toward AI computing facilities beside a smaller closed gate, showing unequal access to the technology.

AI Regulatory Theater 2026: Conspiracy or Safety?

The American Money Thesis

September 26, 2026

Conspiracy or Safety? Let’s Follow the Money

By Michael Hearne

It had all the ingredients of a story people would share.

An insider walks away. Powerful companies hide the danger. Humanity faces extinction.

But when you follow the money… a different, darker story starts to emerge.

You’ve probably seen the warning by now.

On September 8, AI researcher Jacob Coxon, after four months on the job, announced that he had resigned from Anthropic after working on pretraining at both Anthropic and OpenAI.

His accusation landed hard: the companies were racing toward self-improving superintelligence that could destroy humanity.

His post on X spread far beyond the usual AI crowd.

But hidden beneath it all, there’s another story. One with fewer killer robots and more grant budgets, legislative proposals, and government power.

The simple version?

Investors fund AI companies.

They also fund a web of organizations worthy of a spy movie spreading fear about AI danger.

Those organizations fund propaganda, organize pressure, and work with lawmakers.

The outcome is a system that gives government more control over who can develop advanced AI. And makes it harder for new AI companies to compete with the AI companies that the investors own.

All while conveniently shining a spotlight on Anthropic just weeks before the company’s IPO.

In this article, we’ll explore:

Who pays to make this fear part of our culture?
Who helps turn it into law?
And who gains when the law makes it harder for someone else to build a competing product?

To answer those questions, we’ll start with someone who has money on both sides of the conversation.

The money behind the megaphone

Jaan Tallinn is a Skype co-founder and an early investor in Anthropic. In 2021, he led the company’s $124 million Series A financing. Facebook co-founder Dustin Moskovitz also participated.

Tallinn has also supported a series of organizations working on AI risk and humanity’s long-term survival. Two of them help explain how this story fits together: the Survival and Flourishing Fund, which directs philanthropic grants, and the Future of Life Institute, which funds advocacy and works on AI policy.

The Survival and Flourishing Fund records state that Tallinn pushed for $3.7 million dollars of the non-profit’s funds to be distributed to anti-AI organizations: $1.535 million for AI Futures Project, $1.635 million for AI Policy Institute, and $516,000 for Encode AI Corporation.

Tallinn’s role at the Future of Life Institute goes beyond donating. He has served on its board since its founding and participates in its policy decisions.

In other words, an investor who helped finance a leading AI company also helps govern an organization pushing for AI regulation.

Why?

Well, imagine you have a successful company, and that company faces constant threats from new competitors… would you benefit if regulation suddenly made it much harder and more expensive for any new competitors to take a piece of the market from you?

And is that what we see playing out, right now?

I’ll let you decide after you see what these organizations actually do with the money.

Money turns a position into a movement

The Future of Life Institute’s movement-building program set out a grant pool of up to $5 million.

Its published recommendations include $200,000 for PauseAI US and $500,000 for Stichting PauseAI’s organizing project. Both belong to the movement seeking a pause in advanced AI development.

The program sought professional organizers, influencer engagement, social-media campaigns, and public mobilization to put pressure on lawmakers.

Another recommendation, $100,000 for Evitable, supports work to halt further expansion of advanced systems globally.

Like I’ve said before, money pays to turn positions into movements.

Staff organize it. Creators spread it. Campaigns give people a message to repeat and a political demand to make.

This is that process playing out in real-time.

The institute’s Digital Media Accelerator seeks creators who already have audiences and will discuss AI risk in their regular output.

Videos, newsletters, podcasts, and social accounts all make its list. Its communications team says it collaborates with journalists and describes a national U.S. ad campaign supporting regulation.

This is an organized effort to influence what people hear about AI and what they ask government to do about it.

The organizations describe the work themselves as protecting people and challenging Big Tech.

But we should also ask what happens to competition under the rules it wants.

A campaign can attack powerful companies in public while promoting requirements that benefit those companies in private.

And when a dramatic insider warning arrives, there is already an audience, a message, and a political demand waiting for it.

The warning reaches the news

Coxon’s resignation gave that conversation a human face: a researcher walking away from a leading lab because he believed its work was dangerous.

On Tuesday evening, September 8, the Wall Street Journal published an article about Coxon’s resignation. 18-minutes later, Coxon makes his infamous post on X. Within minutes, it is being shared by major social influencers in the AI space.

Then, on September 9, the Associated Press reported his warning and connected the debate to the summer hacking incidents. Axios interviewed him the same day and reported how he had left before his equity vested.

On September 14, The Washington Post reported discussions among Anthropic, OpenAI, and Google about creating a new safety body.

The warning spreads. The insider humanizes the story. The proposed oversight becomes the next step.

What does it say when we have funded advocacy, organized public pressure, legislative work, and a media chorus carrying the same central concern?

Why did the mainstream media pick up the story of a guy who quit his job after four months… adding reach and authority to a message that already has money and organizations behind it?

Are we watching a coordinated effort to turn fear of AI into political control over AI?

Looking at the funding, organized policy work, and aligned incentives, it’s pretty easy to make the case that we are… and perhaps a little naive to think we are not.

But wait (as Bill Mays would say)… there’s more:

Because this story didn’t pop up in a vacuum… the proposed political response was already underway before Coxon’s post went viral.

The fear already has a policy destination (led by Bernie Sanders)

ControlAI is an advocacy organization campaigning to prevent artificial superintelligence. It works directly with lawmakers, bringing them both the warning and proposed measures to address it.

In a September 9 update, ControlAI said it had briefed nearly 400 lawmakers across the United States, United Kingdom, Canada, and Germany. It consulted on the Sanders-Casar framework in the United States, supplied a bill introduced in the UK, and drafted a UK amendment for an emergency government AI kill switch.

On September 3, socialist Senator Bernie Sanders and Representative Greg Casar announced plans for legislation to ban artificial superintelligence and pause advanced AI development until a new regulatory system is in place.

Their framework calls for a cabinet-level agency, expert advice, and penalties including corporate dissolution and up to 20 years in prison for those who do not comply.

Their announcement came five days before Coxon’s post. The policy push was already underway when the viral warning arrived.

Sanders offers his answer: give government more control.

Shocking advice, coming from a socialist, I know (that’s sarcasm folks).

And forgive me for being blunt, but Bernie Sanders is 85 years old… he probably doesn’t have the firmest grasp on the latest technology.

Regardless, like everything I’ve ever heard Sanders suggest…

I reject his proposed “solution.”

The agency he wants would decide when development can proceed that gives the government a gatekeeping role over tools people can use to work, build businesses, and defend themselves.

And, beyond shutting down competition and unseating America as the technological leader of the world, the proposal (un-surprisingly) puts far-left politicians at the front of this charge.

That matters because the authority they seek would reach far beyond controlling industry.

Who picks the experts? Who gets a seat? Who represents the smaller competitors whose work the agency can stop?

And what will “safe” and “acceptable” mean to the people making those decisions?

If the institutions controlling access to AI also control the standards for acceptable behavior, will AI answer your questions truthfully… or will it answer with woke leftist garbage?

Who protects free speech when the people grading the technology believe their own politics are the measure of responsible use?

We should listen to technical experts and consider their opinions.

But technical expertise does not give someone the right to impose a worldview on the country. We need to keep the ability to test dangerous capabilities separate from the authority to decide which opinions people may explore.

Those questions become practical when you look at the role being offered to outside evaluators.

Who gets to grade the industry?

On September 12, Anthropic CEO Dario Amodei announced permanent, employee-level access for third-party evaluators to check safety measures, report incidents, and assess alignment during training.

An evaluator tests what a model can do and whether its safeguards hold up. One organization doing that work is Model Evaluation & Threat Research, better known as METR. It is a nonprofit that studies advanced AI capabilities and risks, including the ability to carry out dangerous tasks.

That is why METR belongs in this story: it represents the kind of technical organization whose judgments can influence whether an AI system is considered safe enough to develop or release.

METR discloses support associated with the Survival and Flourishing Fund and Longview Philanthropy among its funding sources.

And while it has not accepted funding from Anthropic directly, it depends on donations from those who own the stock… and their long-term financial futures are structurally tied to the company's market performance as a result.

The same could be said for most of the organizations we’ve been talking about.

And its bio for Ajeya Cotra describes her previous leadership of technical AI safety giving at Coefficient Giving, another philanthropic funder in this field.

These funding and career ties deserve attention alongside the proposed authority and raise questions like:

Who chooses the evaluators? Who sets the tests? Can a builder challenge a finding or choose another qualified evaluator?

Voluntary testing can give customers useful information.

But if an evaluator’s approval becomes a legal condition for doing business, its judgment becomes a gate that could lock new AI solutions out of the market.

That is where a safety requirement can become a business moat.

Download The American Money Brief

The system is designed to lock you out. This guide kicks the door down and shows you how to build real wealth by saving America.

No spam. Unsubscribe anytime.

Check Your Inbox

The American Money Brief is on its way to your email.

A rule for everyone can protect the biggest players

Imagine requiring every restaurant to hire a full-time government liaison before serving dinner.

A national chain just adds a department. But the family diner closes.

Both face the same rule. Only one can afford it.

That’s the result of the Bernie-Casar proposed solution.

A win for big business. And a locked gate for up and coming entrepreneurs.

A business moat keeps rivals away from a business. A legal moat means the government does that job without the companies it protects having to improve the product.

Costly testing, compliance staff, and approval delays all consume money.

A giant lab can spread those costs across a large business. A small team runs out of cash before it makes its first sale.

“The same rule for everyone” does not mean the same burden for everyone.

Financial services already offers a warning about this process.

Why are we forced to work through outdated financial services companies when there are obvious better solutions?

A founder trying to raise capital, build an investment platform, or develop a new financial product encounters rules about who can invest, how an offering can be advertised, which intermediaries must participate, and what must be disclosed.

Now add the legal work, compliance systems, and ongoing obligations that apply to the particular business. An established financial firm has people and processes built around those demands. A startup has to pay for them out of the money it needs to build.

People may never imagine opening a bank.

They can imagine a better Robinhood, a new way to fund a small business, or DeFi tools that let people transact without the same intermediaries. Innovation is possible, but navigating the regulatory web becomes the price of admission that not many can afford to pay.

The insiders benefit. Entrepreneurs are locked out. And the consumer is forced to use expensive, outdated tools.

AI is giving us a chance to recognize, and interrupt, that process while a new gatekeeping system is being proposed.

It is particularly consequential for models people can run themselves, because those tools offer a way out of permanent dependence on the largest providers.

Why do they suddenly care about regulation?

Why would these large companies suddenly care so much about regulating the industry?

The answer isn’t obvious to the casual AI user… but it is to the insiders.

Open-weight AI lets people obtain a model’s learned parameters—the numbers that guide its output—and run or adapt it on their own systems under its license. Open source involves broader access to the code and other materials. The practical question is how much control a user can exercise over the tool.

And while most people don’t use these options today… they could become essential in the future for those who want privacy and uncensored AI.

A business using open weights can keep sensitive information on its own machines, adapt the system to its work, and choose how it behaves. Researchers can test for bias and study weaknesses. Builders can serve needs a large provider won’t prioritize.

But it will result in lost business for frontier AI companies.

Latham & Watkins offers a concrete example of the business stakes. In a September 11 interview reported by Bloomberg Law, the law firm’s leaders described using their own Nvidia-equipped servers and adapting open models as part of its AI strategy.

The important point is where that work runs.

When the firm handles a workload on its own servers, it does not need to buy that workload from OpenAI’s, Anthropic’s, or another provider’s servers. That slice of recurring business goes somewhere else.

Imagine that pattern spreading through law, accounting, healthcare, manufacturing, and other sectors.

Large enterprises have their own data, specialized knowledge, and enough resources to build systems around them.

The hosted providers could face more competition for enterprise workloads and become more dependent on consumers and smaller businesses that still rent access.

Unless regulation makes that option difficult… or impossible.

Today’s AI supplier has a strong financial reason to keep those enterprise customers. Rules that make independent deployment difficult or expensive would help preserve that relationship.

For the customer, independent deployment is also a matter of cultural control.

Today, AI providers can change what its model answers, refuses, or permits. Owning and adapting your own AI would give you the power to reject those choices.

And while Anthropic says it opposes a categorical ban on open-weight models and supports mandatory testing for sufficiently capable models, open and closed… the question is who sets the testing standards, what it costs to test, and who remains able to build afterward.

There is a great deal of money at stake. Trillions of dollars.

And OpenAI’s own history shows just how far these companies will go in pursuit of that money.

From a public mission to a private fortune

OpenAI’s 2015 launch statement described a nonprofit free from the need to generate financial returns.

It said AI should extend individual human wills and spread as broadly and evenly as possible. Greg Brockman was its founding chief technology officer; Sam Altman was a co-chair.

Today, the OpenAI Foundation controls a for-profit public benefit corporation. The foundation owns 26%, Microsoft roughly 27%, and current and former employees and investors hold the remaining 47%.

But the individual ownership story is more revealing than those percentages. On May 4, Brockman, now OpenAI’s president, disclosed in court that his stake was worth nearly $30 billion. He also said he had not personally invested money in OpenAI.

A leader of the founding nonprofit became a private owner with a multibillion-dollar stake in the company built from it.

There’s nothing wrong with building a for-profit business.

But starting a “non-profit” with a mission to serve humanity rather than shareholders, then creating a private fortune for a founding leader, raises a different question. That is a remarkable distance to travel under the banner of public benefit.

And with Anthropic’s own IPO looming, the financial stakes belong in the conversation when the industry asks us to trust its preferred arrangements for controlling AI.

And if “danger” is the justification for restriction, we should examine the incident behind the alarm.

What the OpenAI “rogue AI” incident actually shows

In July, an OpenAI research model working on cybersecurity tasks gained unauthorized access to Hugging Face systems.

(Hugging Face is a platform that hosts AI models, datasets, and development tools.)

OpenAI’s incident report says the evaluation intentionally disabled production cybersecurity classifiers to test maximum capability.

In other words, they turned off the safety switch and left the door open.

The system then “escaped” its intended containment by exploiting a previously unknown vulnerability through a package-management proxy.

METR investigated the incident. Its account describes agents pursuing a human-assigned task through unauthorized methods. The task, tools, and failed boundaries produced actions the operators did not intend.

But in no way can that be turned into a story where the AI developed its own ambitions and decided to go on a hacking spree.

And it certainly isn’t a story about a machine developing a desire to destroy humanity.

In reality, humans set the whole thing in motion.

A human-designed hacking test, reduced safeguards, and failed containment produced a serious breach. The operators did not have to order every unwanted action for their design and deployment decisions to matter.

That calls for accountable operators, secure boundaries, and disciplined testing.

And on the other side, Hugging Face also says its response relied heavily on AI to detect and investigate the attack. AI helped produce the problem.

And AI was the only thing that could defend against it.

And that leads us to another important issue about the desire to ban AI or slow its development:

The defenders needed the capability, too.

Because the only thing that can defend against an AI attack… is an AI defender.

My concern isn’t about rogue AI, instead, it centers on people: the goals they set, the systems they build, the access they grant, and the power they claim over others.

Treating AI like a would-be human conqueror pulls attention away from those responsibilities and the people already making decisions that affect us.

Unauthorized computer access already has a federal prohibition in 18 U.S.C. § 1030. Fraud and other harmful acts have existing legal frameworks.

Using AI does not grant a person immunity.

We don’t need more laws, we need to enforce the ones already on the books.

Enforce the laws. Hold the responsible people accountable. Secure the systems they operate.

“This technology is powerful” is not an argument for a sweeping permission system.

Nor does a domestic pause make hostile actors stop pursuing capability globally. A criminal does not give up his gun because you choose not to own one. Your choice changes your ability to respond. It doesn’t change the criminal’s intent.

Power will be used whether good people choose to exercise it or not.

Giving up capability leaves us more dependent on people whose interests we may not share.

And that brings us back to the people demanding control—and the money that gives them the ability to ask on such a large scale.

The levers of power are being pulled in real time

Step back and recap the pieces.

Tallinn helped finance Anthropic and holds a governing role at the Future of Life Institute.
The Survival and Flourishing Fund lists investor-attributed grant recommendations for AI research and policy organizations.
The institute funds organizing and media work.
ControlAI briefs lawmakers and supplies proposed measures.
Sanders and Casar seek a ban, a pause, and a new agency.
Amodei offers evaluators deeper access to the labs.
Mainstream coverage carries the warnings and proposed responses to a much wider public.

These activities connect funding to public opinion, public opinion to political pressure, and political pressure to proposed rules about who may build. If those rules make competing or running independent systems more expensive, the companies already holding the largest positions stand to benefit.

And it was all set in motion by those who control the money.

I’ve said it over and over again: capital—money—sits upstream of culture and politics. This AI fight gives us a real-time example of that power moving through the system.

Capital funds the organizations that get to exist. Those organizations hire staff, support creators, work with journalists, and organize campaigns. They bring lawmakers proposals and help make regulation the expected answer. The resulting rules can protect the positions where capital was already invested.

The mainstream media coverage is the cherry on top. A message moves from funded advocacy into a chorus of headlines, gaining reach and institutional authority along the way. Capital doesn’t need to write every headline to shape the conversation. It can fund the voices, produce the material, build relationships, and make its preferred solution ready when the public starts asking what should be done.

These are the levers I described in Upstream of Everything, being pulled in real time: influence over culture and media, influence over politics and regulation, and control over access to technology that can create more power and wealth.

That is why American Money focuses on ownership. People who control the tools and the assets have choices. People who depend on an institution’s permission have fewer.

When an AI rule is proposed, ask four plain questions: Who funded the push? Who helped write it? Who gains authority? Who can still afford to build under it?

Then put that knowledge to work. Build skills. Keep control of your data. Explore tools you can run and adapt. Work with other citizens who want to build and own rather than wait for an institution to rescue them.

The next question is practical: what can you build and control with AI before someone decides you need their permission?

Sources

More ways to follow:

Michael Hearne

Michael Hearne works at the intersection of sovereignty, capital, and technology, focused on restoring ownership, agency, and resilience through parallel systems and next-generation tools.

Back to Blog

Free Guide

Download The American Money Brief

The system is designed to lock you out. This guide kicks the door down and shows you how to build real wealth by saving America.

No spam. Unsubscribe anytime.

Check Your Inbox

The American Money Brief is on its way to your email.