Seven Anthropic Founders Want 50.1 Percent of the Vote on 14 Percent of the Company. Board Elections Are the Carve-Out.

The Information reported on Thursday, September 24, 2026 that Anthropic is asking its shareholders to approve a new class of stock handing its seven co-founders a combined 50.1 percent of the vote on most corporate matters. Reuters and TechCrunch picked it up Friday morning. Anthropic has not commented.

Here is the arithmetic that makes this interesting. Each of the seven co-founders holds roughly 2 percent of the company after more than $85 billion of venture dilution. Seven times two is fourteen. The ask is 50.1. That is a voting leverage ratio of about 3.6x, and the special shares reportedly carry no incremental economic value at all. This is a request for control, priced at zero, with the cash flows left exactly where they were.

I have read a lot of dual-class proposals. This one has a feature I have not seen before, and almost every writeup so far has buried it.

What Is Actually On the Ballot

The structure as reported has five moving parts. The founders get a special class carrying 50.1 percent of the vote on most corporate matters. Control is contingent: at least three of the seven must keep a minimum shareholding, so the bloc survives attrition but not a collapse. Founder board representation rises from two seats to three out of seven, one of which is currently vacant. Employees get their own special class that functions as a tie-breaker on certain questions. And the shareholder vote is expected, in the words of the reporting, in the coming days.

Then there is the exclusion. According to the reporting, the founder class does not vote on electing board members.

The Carve-Out Is the Story

Supervoting stock exists, in almost every case anyone can name, for exactly one purpose: to let insiders seat the board and therefore keep the CEO job. Zuckerberg's Class B is a board-control instrument. Snap's Class C is a board-control instrument. Anthropic is proposing a founder class that reaches most corporate matters and then stops short of the one vote that normally justifies the whole exercise.

That is not a softening. It is a division of labor. Board elections stay where they already are, with the Long-Term Benefit Trust, whose Class T shares carry the right to elect a majority of the seven-person board while the trustees hold no economic stake in the company whatsoever. Layer the two together and you get the actual picture: the founders control most corporate matters, the Trust controls the board, and whoever buys common stock in the IPO controls neither. Two independent locks, and the public float sits outside both of them.

I want to be fair about the read. A founder bloc that cannot seat directors is genuinely less entrenched than one that can, because the body that can fire the CEO is not the body the founders control. But the inverse is also true and less comfortable: the layer holding the most consequential power here is the layer with the least public accountability, no economic exposure, and no shareholder able to vote against it.

How This Compares

Economic and voting figures for Meta, Snap, and Palantir are approximate, drawn from IPO prospectuses and the most recent proxy disclosures; the Anthropic column reflects reported terms of a proposal that has not been voted on or filed publicly.

Anthropic's reported model is closest to Palantir, which pegged its Class F at 49.999999 percent of total voting power at all times regardless of underlying ownership, stopping one rounding error short of a majority. Anthropic is proposing to step over that line by one tenth of a point. The difference between 49.999999 and 50.1 is the difference between very strong influence and a controlled company.

What is actually novel is the group shape. Every structure in that table concentrates control in one person or a pair. Anthropic is proposing a seven-person bloc with a three-person floor, which is a different failure mode. A single-founder class fails when one person leaves. A seven-person class with a three-of-seven condition survives four departures and then breaks all at once.

The Trust That Holds the Board Is Down to Three People

If board elections are the real lock, then the composition of the Long-Term Benefit Trust is the most important governance fact about Anthropic, and it has been moving in one direction. The Trust currently seats three members: chair Neil Buddy Shah, CEO of the Clinton Health Access Initiative; Richard Fontaine, CEO of the Center for a New American Security; and former Federal Reserve chair Ben Bernanke.

A fourth seat has been vacant since August 4, 2026, when Mariano-Florentino Cuéllar left the Trust to become Anthropic's Chief Global Affairs Officer. Read that sequence slowly. A trustee of the body that elects a majority of the board resigned from that body and took an executive role at the company the body oversees, and the seat has sat open for seven weeks while the founders prepared a request for 50.1 percent of everything else.

Three people, no economic stake, one empty chair, and the power to seat four of seven directors at what may be the largest IPO ever priced. That is the governance surface public investors are being asked to buy into.

Why This Week

The timing is not mysterious. Anthropic raised $65 billion at a $965 billion post-money valuation in May 2026. By August the secondary market was marking the company near $1.5 trillion. Reporting suggests the IPO itself may slip past the November US midterm elections. That gap, between a charter amendment you can pass with a private shareholder vote now and an S-1 that becomes a public document later, is exactly the window in which you fix your governance.

There is also a docket now. On September 19, 2026, four paid subscribers filed a Sherman Act class action in the Northern District of California against Anthropic, OpenAI, SpaceXAI, and Google over the pacing accord. A pending antitrust complaint is an S-1 risk factor. So is a controlled-company structure. Getting both settled and disclosed on your own schedule beats amending a charter while the SEC is reading your filing.

Three Counterreads Worth Holding

First, the board carve-out is a real constraint and not a cosmetic one. A founder class that cannot elect directors cannot protect a CEO from a board that wants him gone, which is the single most common use of supervoting stock. Whatever else this structure is, it is not the Meta playbook.

Second, the economics genuinely point away from self-enrichment. The founders pledged in January to give away 80 percent of their wealth, and the supervoting shares carry no extra economic value. If the goal were money, this is a strange way to pursue it.

Third, none of that answers the accountability question. Mission-protective control and unaccountable control are the same instrument described by different people. Every founder-control structure in history was pitched as protection for a long-term vision against short-term markets, and the pitch was sincere in most cases and load-bearing in none of the ones that went wrong.

Our Take

The number that matters is three. Three is the founder floor that keeps the bloc alive, and three is the number of people currently sitting on the Trust that elects the board majority. Two different threes, both load-bearing, both small enough that ordinary attrition is a governance event. A structure that depends on three specific people staying interested is a structure with a personnel risk, not just a charter.

What I find defensible: writing the control structure down, in public, before the S-1 rather than after, at a company whose stated reason for existing is that the technology is dangerous enough to require someone willing to slow it down. What I find harder: the answer to who checks the founders is a three-person trust, the answer to who checks the trust is nobody, and the fourth trustee seat has been empty since the last occupant took a job in the C-suite.

For anyone tracking this as an investment rather than a governance curiosity, the practical read is simple. If this passes, Anthropic goes public as a controlled company under two separate locks, and every index and governance screen that penalizes controlled companies will apply. That is a known, priceable discount. It is not a surprise unless nobody reads the charter.

Three signposts for the next 60 days. Whether the shareholder vote result is published with an actual tally or simply announced as approved, because a lopsided tally and a narrow one say very different things about how the existing investor base feels. Whether the Trust's fourth seat is filled before the S-1 becomes public, because filling it after is a worse look than filling it now. And whether the S-1 risk factors name the Sherman Act complaint and the founder class in the same section, which is the tell for whether Anthropic sees these as one governance story or two unrelated paragraphs. You can track the filings and valuation marks on our Anthropic provider page.