Anthropic and OpenAI agree on almost nothing. One was founded by people who left the other because they thought it was moving too fast. For three years the entire industry has been organized around their disagreement — acceleration on one side, safety on the other, two visions of what artificial intelligence should be allowed to do. On May 30, 2026, the two of them were spending millions of dollars to beat each other in Democratic primaries you have never heard of.

The fight was reported as a fight between super PACs. Anthropic's allied vehicle is called Public First. OpenAI's is called Leading the Future. They are running ads, funding challengers, and working opposite sides of the same midterm races — and a meaningful share of that effort is aimed not at Republicans or at regulators but at each other, in the obscure down-ballot contests where the actual rules for AI will be written.

Two companies whose public identities are built on being opposites turned out to want the same thing badly enough to build the same instrument. Not a better model. A super PAC.

What the labs were built to compete over

The original competition had a clean shape. OpenAI shipped first and shipped fastest — ChatGPT, the developer platform, the deployment everywhere. Anthropic was the counter-position: the lab that would prove you could build frontier models responsibly, that safety was a feature and not a tax. The whole point of having two of them was that they were optimizing for different things. You could tell them apart by what they made and how cautiously they made it.

That distinction has been eroding for a year. The models converged — same benchmarks, same capabilities, same enterprise pitch. The differentiator stopped being the product and became the thing underneath the product: compute, capital, and access. And access, it turns out, is not something you win in a lab. It is something you win in a legislature.

So the competition migrated. It moved from the place where the two labs were genuinely different — their research culture — to the one place where being different no longer helped: the rules. Whatever rules govern AI will govern both of them identically. Which means the only way to compete on rules is to control who writes them.

The phases

The migration did not happen in an announcement. It happened in funding rounds, each one larger than the last, each one looking at the time like the ceiling.

Each stage looked like the equilibrium. A $50M network sounded like the whole game in November. By February the two sides together had raised more than five times that, and the number had stopped being the story. The story was that an industry defined by a disagreement about safety had built two political operations to litigate the disagreement — and pointed them at the same set of primary ballots.

Note which lab is doing what. Public First — the Anthropic-aligned vehicle, the safety side — is the one buying ads and pushing for regulation. The lab whose entire brand is restraint is running the more aggressive political operation. The accelerationists, for their part, are spending to make sure no one slows them down. Both built the same weapon. They are simply firing it in opposite directions, at the same target: the few hundred people who will hold the pen.

The same move, in two other markets

If this were only about elections, it would be a story about lobbying getting expensive. It is not only about elections. On the same day, in two unrelated stories, the same structural thing was happening: capital routing around the open mechanisms that were supposed to govern it.

Anthropic spent May 30 telling the market who is allowed to own its shares. It had named eight secondary-market platforms as unauthorized sellers of its stock; after the notice caused panic and pushback from investors, it cut the list to four. Read that plainly: a private company is policing the market for its own equity, deciding which venues may and may not trade the thing everyone wants to buy. The demand is so far ahead of the supply that the scarce asset is not the product. It is a share of the company that makes it — and you cannot get one on the open market, because there isn't one.

And the Dell contract. The Defense Department's $9.7 billion award to Dell drew scrutiny as a possible return on Michael and Susan Dell's $6.25 billion donation to fund Trump Accounts for 25 million American children.

Dell donation to Trump Accounts
DOD contract that followed

Whatever the truth of the inference, the structure is the same as the super PACs and the same as the share list: the decisive transaction happens off the public ledger. The procurement looks open. The thing that moved it was the donation that preceded it.

In every direction, the open mechanism — the ballot, the share market, the procurement bid — is the surface. The deciding transaction has moved underneath it.

The frontier relocated

The word for what these companies are building used to be "frontier." It meant the edge of capability — the most advanced model, the thing no one else could make yet. That is where the competition was supposed to live, and for a while it did.

The frontier has moved. It is no longer the model; the models converged. It is the interface to the state — the rule that gets written, the contract that gets awarded, the grid connection that gets approved. The same week the super PAC fight surfaced, AI companies were learning to "speak FERC" to fast-track data-center connections to the power grid, and SpaceX won a $4.16 billion contract for Trump's Golden Dome shield. The competition has relocated to the one place where a frontier lab and a defense contractor now compete for the same resource: government attention.

This is the reversal. Two labs that built their identities on being different — on caring about different things, on making different bets about what AI should be — discovered that the difference no longer determines who wins. What determines who wins is the rule, and the rule is the same for both of them. So they stopped competing on what made them different and started competing on the one thing they cannot tell apart: access to the people who write the rules. The disagreement that justified having two labs is now being settled in primaries, by ads, with money.

The earlier reversal — federal rules collapsing into a patchwork of state ceilings — explained where the rules would be made. This one explains who is paying to make them.

The product was never the moat. The moat is the rule. On May 30 the two companies that are supposed to disagree about everything quietly agreed on that — each wiring millions into a primary you have never heard of, to make sure the rule comes out their way.