The AI Cartel No One Is Actually Arresting

The Club No One Asked Them to Run

OpenAI, Anthropic, and Google DeepMind have been holding private meetings about AI safety for weeks. Chris Lehane, OpenAI’s global policy chief, confirmed it on Tuesday in Washington, where he has been lobbying lawmakers on catastrophic risk provisions. The revelation was not a surprise — Bloomberg reported it first, Altman had already hinted at private discussions, and The Information documented plans for a private industry standards body.

What is surprising is how casually the three companies are treating the antitrust landmine beneath their cooperation.

Amodei published an essay Saturday calling for the industry to slow the pace of frontier model development. Altman and Hassabis signed on. Altman went further, promising OpenAI would embed third-party evaluators inside its own organization. Lehane then confirmed the three rivals have been coordinating directly — not through a government framework, not through a legislative mandate, but through private meetings between the most powerful companies in the most powerful sector on earth.

Then came the interesting part: when pressed on whether this coordination violated antitrust law, Lehane said the firms don’t need a government waiver. Amodei’s essay had proposed exactly that — a narrow legal safe harbor for safety collaboration. Lehane dismissed the need. That is either extraordinary confidence or extraordinary recklessness.

The Real Question Isn’t Safety. It’s Moats.

The conventional reading of these talks is straightforward: the biggest AI labs have decided the technology is moving too fast and are trying to hit the brakes together. That narrative is convenient for them. It paints Altman, Amodei, and Hassabis as responsible stewards confronting existential risk while politicians like Donald Trump call safety concerns “a hoax.”

But there is an alternative reading, and it is harder to dismiss.

When three dominant competitors coordinate on technical standards, safety evaluation protocols, and industry pacing agreements — without government oversight — they are not just managing risk. They are managing competition.

Independent verification, the kind OpenAI is now backing through the FRONTIER Act, sounds like transparency. In practice, it is a gatekeeping mechanism. The companies that design the verification criteria, fund the evaluators, and control the certification process get to decide who gets flagged as unsafe and who gets to ship. A startup developing a frontier-class model overnight would face a verification regime designed by and for OpenAI, Anthropic, and Google. The cost of compliance alone would be a moat. The timing of certification — delayed, ambiguous, subject to the judgment of industry-appointed evaluators — would be another.

This is not speculation. It is how standards bodies work everywhere else. The telecom industry did not stumble into compatibility without deliberate coordination. Financial auditing standards did not emerge from a vacuum. When incumbents write the rules, they write them to survive.

Trump’s Blind Spot Is Their Opening

The political context makes this coordination even more consequential. Trump has dismissed AI safety fears. His AI advisor David Sacks — a venture investor with portfolio stakes across the industry — has called existential risk concerns overblown. The administration’s instinct is deregulation and speed, framed as a race against China.

That creates a rare window. For the first time, the U.S. government is effectively outsourcing AI governance to the companies being governed. Hassabis himself called on Washington in July to create a standards body with watchdog powers, but when that never materialized, the three labs simply built their own.

Lehane’s comment that no waiver is needed is telling. It suggests the companies believe they can operate in a regulatory gray zone that Trump’s team will not polices. If antitrust enforcement remains dormant — and there is little signal it won’t be, given the administration’s pro-industry posture — then this private coordination becomes de facto industry governance.

The FRONTIER Act Paradox

Here is where it gets sharper. OpenAI is simultaneously lobbying Congress to pass the FRONTIER Act, which would mandate independent verification for frontier labs. That sounds like accountability. It is also a powerful lever.

If the Act passes, verification becomes law — but the three companies have already been coordinating on what verification looks like in private. They will have shaped the standards before the standards become mandatory. Startups and smaller labs will inherit a system designed by their competitors, enforced by verifiers the incumbents help select, with criteria the incumbents helped write.

The Act also gives the government a ready-made regulatory architecture to plug into. Rather than building its own oversight capacity — which is slow, expensive, and politically fraught — Congress can adopt the private framework the labs have already refined. That benefits the labs in the short term, because the rules favor them. But it also means the government is outsourcing its judgment to the very companies it should be policing.

Who Wins. Who Loses. What Comes Next.

The winners in this scenario are clear: OpenAI, Anthropic, and Google DeepMind. They are coordinating on safety, shaping standards, and positioning themselves as the adult supervisory presence in a room full of chaos. If the government adopts any of their frameworks, they win again — because their frameworks are their frameworks.

The losers are everyone else. Startups building frontier models without the capital to fund third-party evaluations. Foreign labs excluded from U.S.-centric verification regimes. Smaller American labs that wanted to compete on capability and are now asked to compete on compliance.

What happens next depends on two things. First, whether the Department of Justice decides that private coordination among direct competitors on industry-wide standards crosses the line from risk management to restraint of trade. Second, whether Congress treats the FRONTIER Act as genuine oversight or as a rubber stamp for the three labs’ existing arrangements.

If DOJ stays away and Congress passes the Act without independent technical input, the current arrangement hardens into infrastructure. The safety coalition becomes the governing coalition. And the barrier to entry that Lehane insists needs no legal waiver will be enforceable by law.

If DOJ acts, the talks become evidence. If Congress adds genuine independent oversight — not just a mandate for verification but authority over its design — the labs lose their first-mover advantage on governance.

Neither outcome is guaranteed. But the fact that three competitors are holding private safety talks without a waiver, while simultaneously pushing legislation that would make their private standards compulsory, is not a coincidence. It is a strategy. The question is whether anyone outside those three rooms is paying attention to what they are building.