A $2 trillion IPO tests a novel governance structure: a trust with no equity that quietly controls Anthropic's board composition.
Here is the sentence that will ruin an equities analyst’s afternoon somewhere: the most powerful governing body at Anthropic owns none of the company. No shares. No stake in the reported $2 trillion valuation. No line item on the cap table it can be diluted out of.
That body is the Long-Term Benefit Trust, and if you’ve read the coverage you’ve probably absorbed the shorthand that it holds a “safety veto” over model releases. That shorthand is wrong in a way worth understanding, because the truth is both less dramatic and more durable. Let me build it up from nothing, because almost nobody explaining this actually explains the mechanism — they just gesture at it and move on.
Start with the corporate shell, because it isn’t a normal one
Anthropic is a Delaware public benefit corporation. A PBC is an ordinary for-profit company with one bolt-on: its charter names a public purpose, and its directors are legally required to weigh that purpose against shareholder returns when they make decisions. Think of it as a company whose bylaws come with a conscience clause the directors can, in principle, be held to.
In a normal C-corp, a director who leaves money on the table to serve some higher goal is inviting a lawsuit for breaching their duty to shareholders. In a PBC, that same director has statutory cover: balancing profit against the stated mission is the job, not a dereliction of it. Anthropic’s stated purpose is the responsible development of AI for the long-term benefit of humanity. On paper, a director can point at that clause and say “no, not this quarter.”
Now, the honest caveat. PBC duties are famously soft. The balancing test is vague, the case law is thin, and shareholders can’t easily sue a director for being too cautious. The PBC form gives management a shield, not a sword. It’s necessary here, but nowhere near sufficient. Which is exactly why the Trust exists.
The Trust doesn’t veto anything. It picks referees.
The Long-Term Benefit Trust is not a person, a fund, or a regulator. It’s a legal entity that holds a special class of Anthropic stock whose only real power is the right to elect a portion of the board of directors — a portion that grows over time as the company hits pre-set milestones, until the Trust elects a majority of the seats.
Sit with that, because it’s the whole ballgame. The Trust does not review model launches. It does not sign off on safety thresholds line by line. It does not approve or block acquisitions directly. What it does is choose the people who do all of that. It’s the difference between refereeing a match and appointing the referees. The “safety veto” isn’t a big red button someone in the Trust gets to slam — it’s the slow, structural power to make sure the room where releases and M&A get decided keeps a bloc of directors whose job description is the mission rather than the multiple.
This matters for one very practical reason. A veto is a single point of failure — capture the vetoer and you’re done. Board-election power is diffuse and self-renewing. If a trust-elected director drifts or gets bought off, the Trust replaces them at the next cycle. You don’t fight it once; you’d have to fight it every board term, forever.
Who the trustees actually are
The Trust was designed around five seats, filled not by Anthropic executives or investors but by outsiders chosen for judgment on safety, security, and public benefit rather than for their equity or their Rolodex. The initial roster leaned heavily into national security and global health:
- Jason Matheny — CEO of RAND, formerly head of IARPA (the U.S. intelligence community’s research arm) and a White House science-and-tech official. This is the national-security-and-catastrophic-risk seat, and it is not a decorative one.
- Kanika Bahl — CEO of Evidence Action, a global health and development organisation built around measuring whether interventions actually work.
- Neil Buddy Shah — a leader in global health, associated with the Clinton Health Access Initiative. Another operator from the “does this help real people at scale” world.
- Zach Robinson — from the effective-altruism institutional world, the intellectual tradition Anthropic’s founders come out of.
Two things to flag before anyone treats that list as gospel. First, trust membership rotates — seats have already turned over since launch, and the S-1 roster will not necessarily match the announcement roster. Second, notice what’s missing: no famous AI researchers, no ex-regulators-turned-lobbyists, no investors. That’s deliberate. The design bet is that people with no financial upside and a track record in high-stakes, evidence-driven fields will hold a line that insiders wouldn’t. Whether that bet survives contact with a public shareholder base is the actual story here.
What happens when the Trust and the money disagree
This is where the marketing gloss and the legal reality part ways. Because the Trust’s power runs through board composition, a “block” almost never looks like a confrontation. It looks like trust-elected directors voting a certain way, or declining to renew a director who won’t. Slow, procedural, undramatic. That’s a feature.
But Anthropic didn’t build a doomsday device with no off-switch, and the disclosures say so. The Trust’s arrangements include failsafes: its powers phase in over time and against milestones rather than switching on all at once, and the underlying agreements can be amended if a sufficiently large supermajority of stockholders agrees. I’m not going to quote a threshold I can’t verify — the exact percentage is the kind of number people get wrong confidently — but the principle is public. The Trust is not constitutionally unamendable. It’s expensive to amend, deliberately, requiring the kind of shareholder consensus you don’t assemble by accident.
So the resolution mechanism for a genuine standoff isn’t a courtroom or a coin flip. It’s math. Either the mission bloc holds enough board seats to carry the vote, or enough of the ownership base coordinates to change the rules. Everything in between — the ordinary friction of a launch someone wants slowed — gets absorbed inside the boardroom, out of public view.
Why this isn’t OpenAI’s structure with a new coat of paint
The lazy comparison is “both companies bolted a mission onto a for-profit.” The mechanisms are genuinely different, and the difference is instructive.
OpenAI’s original design was a capped-profit LP sitting under a controlling nonprofit. Investor returns were capped (early rounds reportedly at up to 100x, declining for later money), and the nonprofit board directly governed the whole thing — it could hire and fire the CEO, which it famously did and then undid over a single weekend in November 2023. That episode is the clearest stress test any of these structures has had, and the lesson cut both ways: the nonprofit had real teeth, and it also nearly ate the company, then blinked. OpenAI has since moved toward a PBC restructuring with the nonprofit retaining control and a large equity stake, unwinding the profit cap in the process.
Compare the levers. OpenAI’s controller holds the purse and governs directly — concentrated, powerful, and, as we saw, brittle under pressure. Anthropic’s Trust holds no purse and governs indirectly, by appointing referees, and can ultimately be overridden by a large enough shareholder supermajority. One is a single strong hand on the wheel. The other is a rulebook about who gets to touch the wheel. Neither is obviously better. They fail in different ways.
The test that actually starts now
Here’s what an IPO changes, and it isn’t the valuation. Every governance structure above looks fine when the shareholders are mission-aligned venture funds who signed up for exactly this. Public markets don’t care about your charter. They care about the number, every ninety days, and they have well-worn tools for applying pressure: proxy fights, activist stakes, campaigns to expand or replace the board.
The mission bloc’s power depends on holding board seats. Public shareholders’ entire playbook is influencing board seats. That’s not a hypothetical collision — it’s the same lever, pulled from opposite ends. The PBC shield lets directors say “we balanced mission against profit and chose this.” It has never been tested by a shareholder base that wants the profit part to grow every quarter and has lawyers on retainer.
Watch the boring stuff, not the dramatic stuff. Nobody will announce a blocked model launch. The tell will be in the proxy statements — how many seats the Trust still elects after the lockups expire, how the amendment thresholds hold, who quietly leaves the board and who replaces them. The safety veto everyone keeps writing about was never a button. It’s an org chart. And org charts are exactly what public shareholders are best at rewriting.