Who actually controls Anthropic
In plain English
- The claim is close. Four people do sit on the body that elects four of Anthropic's seven directors. But four is not the whole story, and they are not the only key.
- Control needs two keys. The four trustees hold one share of Class T stock and elect four board seats. Seven co-founders hold one share of Class F stock that casts 50.1% of every shareholder vote and elects the other three seats. Neither block can remove the other's directors.
- Shareholders own the company and cannot steer it. Buy ANT and you get one vote per share, a permanent ceiling of 49.9% on all Class A votes combined, and no director you can choose on your own. Amazon and Google will own about 28% of the company, in shares that cannot vote at all.
- Neither key expires on a calendar. Founder control ends only when fewer than three of the seven founders remain. Trust control has no expiry date; ending it takes a supermajority amendment. Common shareholders get control last, if ever.
- The uncomfortable part. The senior key belongs to a body that selects its own members, owns no equity, and answers to nobody who can vote it out. That is new for a company this size.
The short answer
Anthropic listed on Nasdaq this morning as ANT, after pricing the largest IPO on record: 980 million Class A shares at $83, raising $81.3 billion at a fully diluted valuation of about $1.5 trillion. The stock opened at $111.50, 34% above the offer price, and was trading around $95 by early afternoon.1
The structure that matters was set before any of those shares existed. Anthropic's charter creates five classes of common stock. Two of them are single shares with no economic rights, and between them they decide who sits on the board.2
Figure 1 — The two keys. The Founder LLC's Class F share and the trust's Class T share are the only instruments that elect directors. Public Class A shares vote, but cannot outvote 50.1%.
So the version of the story that says "four people control Anthropic" is right about the board and wrong about the company. Four trustees elect four directors: a majority of the seven-seat board, and enough to decide who runs the place, including whether the CEO stays. But the same charter gives seven founders 50.1% of every shareholder vote and the other three board seats. Four people control the board. Seven people control the shareholder vote. Nobody controls both.
That is not a loophole. It is the design: two independent power centres, each able to block the other, neither able to be removed by the other. The filing says it plainly: "Collectively, the LTBT and the Founder LLC will have the ability to elect all members of our board of directors."2
The trust: four people, one share, no money
The Long-Term Benefit Trust was written into Anthropic's charter at the close of its Series C in 2023. The company described it then as "an independent body of five financially disinterested members with an authority to select and remove a portion of our Board that will grow over time (ultimately, a majority of our Board)."3
It is a Delaware purpose trust. It holds one share of Class T common stock — after a 1-for-1,000 reverse split in September, still one share — which carries no dividends, no liquidation rights beyond par value, and no economic interest in Anthropic at all. What it carries is the right to elect and remove four of the seven directors.2
| Trustee | Since | Background |
|---|---|---|
| Neil "Buddy" Shah (chair) | June 2023 | CEO, Clinton Health Access Initiative; formerly GiveWell |
| Richard Fontaine | April 2025 | CEO, Center for a New American Security |
| Ben Bernanke | June 2026 | Former Chair, US Federal Reserve; Brookings |
| Christopher Stubbs | October 2026 | Experimental physicist, Harvard |
| Fifth seat | — | Vacant |
Two properties make this body unusual, and both are in the filing rather than in a critic's characterization.
It selects its own successors. "The Trustees have the authority to appoint the Trustees by majority vote of the Trustees then in office," with 30 days' notice to the board's nominating committee, a notice period the committee can waive.2 The four current trustees will choose the fifth. No shareholder votes on any of them. Anthropic's original announcement says its board chose the initial trustees; the filing adds that "in practice to date, the Trustees and company executives have collaborated closely on sourcing, interviewing, and selecting Trustees."2 That is an honest sentence, and it tells you the independence is structural rather than adversarial.
It has no expiry date. The filing's risk factors say the Class T stock "will remain outstanding indefinitely," subject only to amendment or redemption through a threshold that requires either 80% of the board plus 80% of the trustees, or supermajorities of the founders, employees and Class A holders acting together.2 There is no date on which the trust's power lapses.
The boundary is also porous in at least one direction. Tino Cuéllar left the trust in August 2026 to become Anthropic's Chief Global Affairs Officer — a trustee moving into management, which the company discloses in a footnote to its own governance page.3
Figure 2 — The seven seats after the offering. The trust fills four (three named plus a vacancy); the Founder LLC and Class A holders together fill three. The trust's four seats are a majority of the board.
The founders: seven people, one share, 50.1%
The second key is simpler and, in one respect, stranger. Anthropic issued a single share of Class F common stock to a new LLC whose only members are the seven co-founders: Dario Amodei, Daniela Amodei, Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish and Chris Olah.2
That one share is entitled to 50.1% of the total voting power of the company on every matter submitted to shareholders, including the election of the three directors the trust does not pick. It cannot be diluted: the charter fixes it at 50.1% of the vote on each record date, however many shares the company issues later. It has no economic rights, pays no dividend, and is not transferable except in a reorganisation that does not change the LLC's membership.2
Inside the LLC, each founder gets one vote regardless of how much stock he or she owns. A majority decides how the Class F share votes. If there is no majority, the share votes in proportion to the Class A vote — the one path by which public holders could, briefly, matter.2
Ownership is not the point, but there is a floor. Each founder is removed from the LLC if he or she stops holding at least 2% of the shares held at the IPO. Since each co-founder has pledged to give away 80% of his or her equity, that floor is the thing that keeps the vote attached to the person rather than the shares. It is also why the founders can give away most of the money and keep all of the control.2
And the control has no calendar end. Several summaries say founder control expires ten years after the IPO. The filing does not say that: the tenth anniversary is when successor members stop counting toward the three-founder minimum. The original seven can hold the Class F share indefinitely. It is redeemed only when fewer than three Continuing Founders remain for 90 consecutive days, followed by a 450-day transition period — or if a supermajority of the founders votes to give it up.2
What a shareholder actually buys
Class A stock gets one vote per share. It just cannot win. Because Class F is fixed at 50.1% of total voting power, all Class A shares together — the public's, the founders', every employee's — are capped at 49.9%. The filing states the consequence without decoration: Class A holders "generally will not be able to affect the outcome of any matter submitted to a vote of our stockholders or the election of our directors, regardless of how they vote."2
The three directors that Class A votes on are elected by Class A and Class F voting together, which means the founders decide them. So a public holder cannot choose a single board seat, cannot remove a director, and cannot force a vote on a merger.
Figure 3 — Equity and votes. Amazon and Google hold 28.4% of the company between them, all in Class N shares with no vote. The founders hold 12.8% of the equity and 50.1% of the votes.
The largest economic holders are the clearest illustration. Amazon will hold 14.7% of Anthropic's capital stock and Google 13.7% after the IPO and the concurrent $26 billion private placement — 28.4% combined — and none of it votes. Both hold Class N shares, which carry the same economics as Class A and no vote, and convert to Class A only when sold to an unaffiliated buyer.4
Amazon also kept convertible notes that would take it to 23.8% if converted in full; it has not waived the ownership cap that keeps them outstanding. The rest of the $26B placement: Nvidia $10B, AMD $2.5B, Broadcom, Micron and Microsoft $1B each, and TSMC Partners $500M.4
There is also a charitable block inside the company: 846,840,960 Class N shares, about 5% of fully diluted shares as of June 30, 2026, sit in the Anthropic Public Benefit Fund, a wholly owned subsidiary. The board has approved issuing up to another 2% each time the share price rises 50% above the prior milestone, starting from the IPO price.2
Shareholders do have one lever, and it is a courthouse rather than a ballot box. Delaware's Public Benefit Corporation statute lets stockholders sue to enforce the board's duty to balance profit, mission and the people affected by the company — and for a listed company the standing threshold is the lesser of 2% of shares or $2 million in market value.5 That is a real check on conduct. It is not control.
The one check on the trust
The structure is not quite "the trust does whatever it wants." One of its four seats — the "Specified Class T Director," currently Chris Liddell — has a procedural check that runs through the founders and then the employees.2
If the trust wants to nominate, replace, or remove the Specified director without cause, the board's nominating committee must concur. That committee has three members: one Class T director and two directors who are not Class T directors — in practice, two elected by the founder-controlled vote. If the committee refuses, the question goes to the holders of Class E stock: employees.2
Class E is the elegant part. Employees receive one share for every six months of employment, plus one share for each share they pledge to charity through the company's matching programme. Class E has no economic rights and votes on exactly two things: a fight over the Specified director, and amendments to the trust itself. Shares are redeemed when people leave, and all of them are redeemed when Class F ends.2
That is the OpenAI November 2023 lesson written into a charter: if the board and the mission guardians disagree, the employees break the tie. It is a genuine check, but a narrow one. It covers one of the trust's four seats, and only nomination or removal without cause. The other three seats are the trust's alone. As Harvard law professor Jesse Fried put it to the New York Times, the founders' new class "watered down" the trust's authority, though it did not "fundamentally change the governance of the company."6 The lower panel of Figure 2 traces the path.
Can the trust do anything it wants?
No. It is the strongest single bloc, but it is one of two, and the charter is built so that the decisions that change the company need both. The table is the whole argument.
| Decision | Trust alone? | What it actually needs |
|---|---|---|
| Elect or remove its four directors | Yes | Three at will; the Specified seat runs through the nominating committee and, if refused, an employee vote |
| Hire or fire the CEO, set strategy, approve a budget | Yes | Four of seven board votes — a majority |
| Remove Dario, Daniela or Yasmin Razavi | No | Only the Founder LLC and Class A holders can; the trust has no vote on those seats |
| Approve a sale or merger | No | Board approval and a shareholder vote; Class F is 50.1% |
| Amend the charter | No | Shareholder vote, where Class F is 50.1% — and amendments to Class T rights need the trust's consent too |
| Amend or dissolve the trust | No | 80% of the board (6 of 7, so at least two founder-elected directors) plus 80% of trustees, or founder + employee + Class A supermajorities |
| Change the size of the board | No | 80% of the board — the seven-seat structure is locked by both blocs |
| Pay itself | No | Class T pays no dividend and has no liquidation rights beyond par value |
Two consequences are easy to miss. The first is that the trust cannot instruct its directors. It can appoint and remove them; it cannot tell them how to vote. They are fiduciaries with a duty to balance profit, mission and the people affected by the company, and the trust's power is the blunt lever of removal, not a remote control.
The second is that the trust is not sovereign even over itself. Three of the five authorized trustees can remove a trustee, and the four current trustees choose the fifth. It is an oligarchy with a succession rule, not a single ruler.
The closest thing to absolute power in the structure is the CEO's job. Four board votes can end it. That is the sharpest fact here: four trustees elect the majority that can fire the chief executive of a $1.5 trillion company, and no shareholder can remove them for doing it.
How few people could control it?
A useful stress test is to stop counting shares and count people. The charter and the trust agreement make that easy, because almost every threshold is written as a number of human beings.
| To control… | People | Mechanism |
|---|---|---|
| The trust's four board seats | 3 | A majority of the trustees then in office — three of the four seated today |
| The trust itself, permanently | 3 | Three of the five authorized trustees can remove a trustee and appoint the successors |
| The Founder LLC's 50.1% | 4 | One vote per founder; a majority of the seven decides how Class F votes |
| The board's ordinary decisions | 4 | Four of seven directors — or three trustees, through the four directors they appoint |
| A sale or a charter change | 7 | Three trustees for the board, plus four founders for 50.1% of the shareholder vote |
Two numbers fall out. Three is the smallest group that can end a chief executive's tenure. Seven is the smallest group that can sell the company. Both are far below the number of people who would have to agree at a normal public company this size — which is the point of the design, and also its cost.
The trust agreement puts a floor under the first number. The trustees cannot exercise any power of the Class T share without a majority of those in office, and they cannot act at all if fewer than three have been in office for more than ninety days. A two-person rump is not allowed.7 One exception runs the other way: a defined set of "Specified Matters" needs 80% of the trustees, which with four seats means all four. On those, no three-person majority is enough.
Coercion is the wrong model for this, and not only because it is a felony. The trustees own no equity, take no share of profits, and cannot sell their position — there is no financial leverage to apply. A compromised trustee can be removed by the other three. Directors are fiduciaries, so a vote cast under duress is voidable and the disclosure and criminal exposure are severe. The founders' share is voted by an LLC majority, so compromising one founder changes nothing. The structure is not hardened against a determined state actor; nothing this small is. It is unusually resistant to the ordinary levers.
The cost is the mirror image of the benefit. A body of four people is cheap to convene and easy to understand, and the integrity of three individuals becomes a systemic risk for a $1.5 trillion company. That is the trade the charter makes: it takes the company out of the hands of millions of shareholders and places it in the judgment of a handful of people who own nothing.
The capture problem: three people is a small attack surface
If the premise of this post is right — that a handful of people hold a decisive lever over the most consequential technology of the century — then the interesting question is not who they are. It is who else has noticed, and what they would do about it.
Control of Anthropic is not control of the world. It is control of one of a few frontier labs, run by thousands of people and constrained by customers, regulators, capital and physics. But if the first system to cross a threshold shapes everything after it, then this is the largest non-state lever anyone has ever held. Actors who think in decades will try to hold it. Some will start with money, some with law, some with people.
Blackmail is the obvious attack, and the least efficient one here.
- There is nothing to buy. The trustees own no equity, take no share of profits and cannot sell their position. The usual levers — debt, stock, a payout — have nothing to attach to.
- A compromised trustee is removable. Three of the five authorized trustees can remove a trustee, and the other trustees choose the replacement.
- A coerced director's vote is voidable. Directors are fiduciaries; duress voids the act and is a crime for the person applying it; and a public company leaves a paper trail.
- The founders' share is one LLC share voted by a majority of seven. Compromising one founder changes nothing.
But the check fails at three. The same majority that lets the trust act lets it expel the fourth trustee. Three compromised trustees are not a scandal to be removed; they are the trust. The removal power is a defence against one bad actor, not against a coalition. A body of four is below the number at which most institutions holding this much power require dual control.
The efficient attacks are slower and legal.
| Attack | Cost | Does it work? |
|---|---|---|
| Buy influence | Money | No. There is no equity to buy and no profit share to divert. |
| Compromise one person | Personal leverage | Partly. One trustee can be removed; three cannot. |
| Capture an appointment | Years, and legal | Yes. Four trustees choose the fifth; the vacant board seat is filled by the trust; the Specified seat runs through a committee. |
| Legal and political pressure | State power | Yes. Subpoenas, investigations, export controls, procurement and national-security review need no compromised individual. |
| Shape the successor pool | A decade | Yes. Trustees serve one-year terms with no limit, and choose their own successors. |
The cheapest path is not a secret. It is the fifth trustee. With four trustees, a 2–2 split has no majority; the fifth seat breaks the tie, and a trustee can serve indefinitely. An actor who cares about 2040 does not need to compromise anyone today. It needs to be helpful to the right four people for a decade, and then help choose the fifth.
That is the structural cost of the design. A public company disperses capture risk across millions of holders, a board that can be replaced and a proxy fight that happens in daylight. This charter concentrates it into four people and calls it independence. The trust's financial disinterest removes the cheapest lever — and also removes the signal. A trustee being paid leaves a trace in a disclosure. A trustee being pressured does not.
What hardening would look like. Publish the trustee selection criteria and each trustee's outside income and foreign contacts. Extend the employee Class E check from the Specified seat to all four trust seats. Require 80% of trustees, not a bare majority, to elect or remove a director. Commission an independent annual audit of the trust with a public report. Adopt a capture protocol — disclosure, recusal, suspension and a pre-agreed replacement path — before it is needed. Add term limits and a cooling-off period before a trustee can take a company role; one already moved from the trust to the executive team in August.3 And for a company that treats national security as a core risk, treat trustee appointments as security-relevant appointments.
None of that is in the charter. The structure was built to protect the mission from shareholders, and it does. The price is that it protects the mission with three people. If AI is as consequential as the company says it is, that is not a detail of the governance. It is the governance.
What would change this
Figure 4 — What ends when. Founder control has conditions but no date; trust control has no date and only a supermajority amendment. Public shareholders inherit control only after both are gone.
- The fifth trustee. The four current trustees choose the fifth. Watch who they pick; it is the only appointment that changes the trust's character without any shareholder vote.
- The vacant board seat. The trust fills it. When it does, the trust's bloc goes from three of six seated directors to four of seven.
- A founder leaving. Control is a headcount now. Fewer than three remaining founders for 90 days starts a 450-day clock, after which Class F and Class E are redeemed and the founders' 50.1% disappears.
- Class N conversions. Amazon's notes and any sale of Class N shares to unaffiliated buyers convert into voting Class A. That does not dilute Class F — it is fixed at 50.1% — but it changes who owns the 49.9%.
- An amendment. With the trustees' consent, 80% of the board plus 80% of trustees can change the trust. Without it, the founders, employees and Class A holders need simultaneous supermajorities. After Class F ends, 85% of Class A can end the trust.
Bottom line
The IPO sold economics, not control, and the buyers knew it. The prospectus says so in the risk factors, in the summary, and in the capital-stock table. What is new is not that founders kept the vote — Musk did it at SpaceX in June, Zuckerberg did it at Facebook — but that the senior key belongs to a self-perpetuating body with no equity, no constituents, and no expiry date. Four people elect the majority that can replace the chief executive of a $1.5 trillion public company. No shareholder can replace the four people.
Sources
- Anthropic, pricing release, 18 November 2026; Reuters/AP coverage of the 19 November open; live quote from stockanalysis.com at 1:09 p.m. ET.
- Anthropic, PBC, Form S-1 (26 October 2026), including "Our Governance Structure and Its Goals," "Management — Board of Directors," "Description of Capital Stock," and the beneficial-ownership table; Amendment No. 3 (17 November 2026) for Class N holdings.
- Anthropic, The Long-Term Benefit Trust and company governance page, accessed 19 November 2026.
- TechCrunch, Amazon and Google will together own nearly 30% of Anthropic after its IPO, 17 November 2026; S-1/A No. 3 share counts.
- Delaware Code, Title 8, § 367, Suits to enforce § 365(a).
- New York Times, After Anthropic's public debut, will questions be asked about its unique structure?, 18 November 2026; Fortune, Anthropic's dual-class share structure, 26 October 2026; TechCrunch, Who's really in charge at Anthropic after the IPO?, 27 October 2026.
- Anthropic Long-Term Benefit Trust, Amended and Restated Trust Agreement (Exhibit 99.1 to the S-1), § 2.1(e)–(f).