The Majority of the Board Belongs to a Trust
This post is part of my Medium blog.
Anthropic filed its S-1 this morning. Class A shares, Nasdaq, ticker ANT. Goldman Sachs and Morgan Stanley at the top of a very long list of underwriters. The numbers are what you would expect from the company that sells Claude: $4.6 billion of revenue last year, $16.3 billion in the first half of this one, revenue per employee north of $6.6 million.
Those are the numbers everyone will talk about today. They are not the reason I read the thing.
The S-1 describes a company where the majority of the board is not elected by the people who own the money. Anthropic is a public benefit corporation under Delaware law, which means its directors are legally required to balance shareholder returns against the interests of "those materially affected by our conduct" and a written purpose in the charter: the responsible development and maintenance of advanced AI for the long-term benefit of humanity. That clause is now in a document filed with the SEC.
Then it gets stranger. A body called the Long-Term Benefit Trust — five trustees, "financially disinterested," chosen for expertise in AI safety, national security, public policy, and social enterprise — holds a class of stock, Class T, that elects and removes the Class T directors. Those directors together are a majority of the authorized board. Not the founders, in that sentence. Not the investors. The trust.
The founders did keep something. A single share of Class F common stock, held by an LLC owned by the seven co-founders, carries 50.1 percent of the vote on anything shareholders get to vote on. One share, half the room. And Class A, the stock you could actually buy, gets one vote per share — and elects almost nobody.
So the structure is: a trust picks the board, the founders out-vote everyone on shareholder matters, and the public buys the economic exposure without the control. Anthropic says the quiet part out loud in the risk factors, as these filings do. They consider themselves a "controlled company" under Nasdaq rules. They mean it literally.
I keep looking at that trust and thinking about what it is for. The S-1's own language is that the LTBT is "designed to supply both the incentives and insights for striking this balance appropriately." Strip the legalese and the bet is: five people with no money in the game will make better calls about advanced AI than the market would. Maybe. It depends entirely on who picks the trustees, and after this offering, the answer to that is: the trust's own succession process, seeded by the founders. Which is to say, the founders chose the people who will choose the people who pick the board.
That is not a scandal. It is a design. It is also the exact shape of the thing I keep writing about.
In Essential, the third book, the question that runs underneath every chapter is who holds the pen when a system starts making decisions people have to live with. The answer in the book is never "nobody." There is always somebody, and half the tension is finding out who. The anthill of governance — charters, boards, trustees, oversight bodies — exists to make that finding-out happen before the damage instead of after. Anthropic has now written that question into its own certificate of incorporation and filed it publicly, which is more than most companies building this technology have done.
And still. A trust with no economic rights is only as strong as its trustees' independence, which is only as strong as the selection process, which the founders control at the seed level. When I wrote the governance scenes in the trilogy, I gave the oversight bodies teeth and then made the teeth a problem of their own. Power that cannot be bought can still be parked. A disinterested trustee is not the same thing as an accountable one, and nothing in an S-1 makes a person accountable.
The other thing the filing makes plain is the money. Anthropic has raised $141.1 billion and holds $112.4 billion in cash. That is not a company. That is a research program with a nation-state's treasury attached, going to the public markets for more. The S-1 argues the two loops — better models, more investment — reinforce each other, and the revenue numbers back them up. I don't doubt the loops. I read a lot of filings and I have learned what compounding looks like on a page.
But the offering's real product, the thing being listed, is not Claude's API revenue. It is a stake in the answer to: who decides. Anthropic has answered in advance, in public, at the level of corporate structure rather than press release. The trust decides the board. The founders decide the shareholders. The public gets to watch, and to be materially affected.
There is a version of this where the trust is a costume the founders wear to keep control while looking responsible. There is a version where it is a genuine hand on the wheel, held by people with no stake in the speed. The document cannot tell you which one it is. Only the trustees' first hard call will, and we will not get to read that filing.