George Stigler won a Nobel Prize for an insight so simple it sounds like a joke: industries ask for regulation because regulation helps them.
The naive view is that regulators protect the public from industry. Stigler showed the opposite is more common. The industry captures the regulator. Rules get written to benefit the people being regulated. The public interest is the cover story. The private interest is the substance.
This is called regulatory capture. It is one of the most well documented phenomena in political economy. And it is happening right now in plain sight, on the front page of every newspaper, with the full approval of the people who will benefit.
The biggest AI labs in the world are asking to be regulated. They are not asking quietly. They are publishing manifestos, coordinating with each other, and spending hundreds of millions of dollars on lobbying to make sure the rules go their way.
You are supposed to believe this is about safety. Let us look at what is actually happening.
The Cartel Speaks With One Voice
On July 16, 2026, Axios published a story with a remarkable headline: “Behind the Curtain: AI godfathers converge on regulations."
The three men racing to build superhuman AI - Demis Hassabis of Google DeepMind, Sam Altman of OpenAI, and Dario Amodei of Anthropic - all published detailed regulatory proposals within weeks of each other. All three agreed on the core architecture: independent testing before release, a single governing system, and U.S. leadership.
Anthropic’s Jack Clark put it bluntly: “At this point, everyone at the frontier of AI agrees that third parties should test out AI systems and use these to develop standards to feed into policy."
They disagree on the details - Amodei wants an FAA-style agency with immediate power to block models, Hassabis wants an industry-funded FINRA-style body, Altman wants an IAEA-style international forum. But the direction is identical.
All three want licensing regimes. All three want government oversight. All three want the state to decide who can build frontier AI and who cannot.
And all three of their companies have the armies of lawyers, the government affairs teams, and the balance sheets to navigate whatever certification process gets created.
The same Axios article noted the obvious implication: “OpenAI, Google and Anthropic already have the lawyers, security teams, government relationships and technical staff to navigate a complex certification process. Startups and open-source developers would face a much steeper climb."
The Compliance Cost Is the Moat
Ask yourself what a “safety regime” for frontier AI actually requires.
You need a dedicated safety research team. You need red-teamers who can probe your models for catastrophic risks. You need compliance officers to document every training run, every evaluation, every deployment decision. You need lawyers to handle the licensing applications. You need government relations staff to manage the relationship with the regulator. You need enough cash reserves to survive a months-long certification process before you can release a product.
A startup with three engineers and a breakthrough idea cannot do any of this. A big lab with billions in funding can.
This is the oldest trick in the regulatory playbook. The call for “high standards” sounds noble. The reality is that high standards are expensive. The only companies that can afford them are the ones already in the market.
OpenAI and Anthropic each spent over $1 million on lobbying in Q1 2026 alone, according to OpenLobby’s tracking. Anthropic posted $1.6 million - its biggest quarter ever. Combined, the two AI developers spent $3.17 million in Q2 2026, up 23% from Q1. Total federal lobbying hit $5.3 billion in 2025, and 2026 is on pace to break that record.
Forbes put it perfectly in February 2026: “AI’s biggest builders are now its biggest lobbyists."
The companies building the most powerful AI systems are spending the most to influence how those systems are regulated. They are simultaneously writing the technology and writing the rules. If you cannot see the conflict of interest, you are not looking.
At the Axios AI+NY Summit on June 8, 2026, startups warned exactly this: flawed regulations could hand the industry’s future to dominant giants before small competitors can push back. The people closest to the problem can see what is coming. They are the ones who will be crushed.
The Energy Wall Changes the Calculation
There is a second reason the big labs are suddenly keen on regulation. Sabine Hossenfelder explained it in May 2026: AI is about to hit an energy wall.
The scaling approach that drove every major AI advance - more data, more compute, more parameters - is running into physical limits. Training frontier models requires staggering amounts of electricity. The cost of compute is not falling fast enough. The returns to additional scale are diminishing.
As one Reddit observer put it: “The Energy wall just means the concept of scaling to achieve AGI has failed, now they need to pivot to AI governance.”
If you cannot build the future, the next best thing is to regulate who can try.
The energy problem is not theoretical. Goldman Sachs projects U.S. data center power demand will rise from 31 GW in 2025 to 41 GW in 2026 and 66 GW in 2027. That is a doubling in two years. Data centers will consume 8.5% of total U.S. power by 2027.
And the public is starting to push back. Hard.
The Brookings Institution reported on July 7, 2026 that local citizen outrage blocked or delayed 75 data center projects worth $130 billion in the first three months of 2026. That is roughly the same number as all 12 months of 2025 combined.
Wisconsin’s Dane County imposed an 18-month moratorium on hyperscale data centers. Maine proposed a pause until 2027. Eleven other states considered similar measures. The backlash is real, it is bipartisan, and it is accelerating.
“The debate over land use also represents the crystallization of broader public concern about AI’s effect on American lives and livelihood,” Brookings wrote. “Americans don’t know how to fight AI. So they’re fighting data centers.”
The big labs see this. They know their expansion plans are generating political resistance. Regulation offers a way to manage that resistance - to channel it through a process they control, toward outcomes that protect their incumbency.
The Hyenas Circle
This is the pattern we have seen before. As I wrote in “The Grievance Machine, Part II”, the hyena dynamic operates through aligned incentives. Nobody has to plan the capture consciously. The incentives align so perfectly that conscious planning is unnecessary.
Politicians need a crisis to solve. AI is the crisis of the moment. Regulating it wins votes.
Bureaucrats need a problem to manage. A new AI agency means new budgets, new staff, new prestige. Every report from that agency will conclude that more authority is needed.
Journalists need a story. “AI safety rules protect the public” is a better headline than “AI safety rules protect Google’s market share.”
And the companies themselves need a moat. Regulatory compliance is the most durable moat ever invented. It costs hundreds of millions, requires armies of lawyers, and takes years to navigate. A startup with a better model cannot cross it. A competitor in another country cannot cross it either.
The same pipeline that funnels taxpayer money through quangos to produce “independent” research - the sock puppet industry I documented earlier - is the same pipeline that will produce the academic cover for AI regulation. The grants will go to researchers who confirm the danger. The think tanks will produce the reports that justify the licensing regimes. The media will amplify the conclusions. The politicians will act.
And the three companies paying for it all will be the only ones left standing when the rules are final.
The Question You Have to Answer
Let me be clear about what I am not saying.
I am not saying AI poses no risks. It might. But the people telling you about those risks are the same people who stand to profit most from the regulations they are proposing. That is not a conspiracy. It is an incentive structure. And incentive structures are more reliable than good intentions.
I am not saying regulation is always bad. Some rules are necessary. But the regulatory framework should not be written by the people who will benefit from it. That is not safety. That is capture.
And I am not saying the three labs are acting in bad faith. They might genuinely believe their own safety arguments. But belief does not change the structural outcome. The effect of their proposals is to entrench themselves and exclude everyone else. That effect is independent of their intentions.
The question is simple.
Do you want the rules for AI written by the three companies that dominate AI, the regulators they helped appoint, and the experts they fund?
Or do you want a system where anyone with a good idea can compete, where the barriers to entry are technical and not bureaucratic, where safety is demonstrated through competition rather than decreed through licensing?
The Wild West era of AI development is ending. The question is what replaces it.
The big labs have already published their answer. You should read it carefully. The safety language is the cover. The moat is the substance.
Are you going to let them build it?