The Governance Experiment at the Center of Anthropic's IPO

The document that will determine whether Anthropic's governance experiment survives is still sealed.

The company — which operates as a public benefit corporation governed by something called the Long-Term Benefit Trust — filed its S-1 with the SEC on Monday. But the filing that matters is not the one that went public. It is the stack of governance exhibits underneath it: the LTBT charter, the trustee appointment procedures, the override rights, the list of who actually sits on the body charged with keeping Anthropic aligned with its stated mission. Those documents will not surface until Anthropic chooses to release them — or until 21 days before a roadshow, whichever comes first.

What is known now is that only three of the trust's five trustee seats are filled. The trust cannot function as designed with a permanent three-to-zero majority on every decision that matters. Until two more trustees are named and seated, the body meant to hold Anthropic to its mission operates under a structure that cannot be overruled.

Anthropic has not disclosed who the three current trustees are, what decisions they have made, or whether they have ever overridden a leadership choice on safety grounds.

The LTBT was designed to solve a specific problem. Anthropic's founders, most notably Dario Amodei, wanted a structure that would prevent the company from making the tradeoffs that commercial pressure typically forces on AI labs — cutting safety investment to boost short-term revenue, for example, or ignoring alignment research when a competitor ships first. The trust gives a board elected by its trustees the authority to override management decisions that threaten the company's stated mission of responsible AI development.

The design has a known vulnerability. Harvard Law Review analyzed it in March 2025: the LTBT was built to constrain traditional shareholders, but Anthropic's two largest investors — Google and Amazon — are not traditional shareholders. They are what scholars call "superstakeholders": entities with strategic interests in Anthropic's behavior that extend far beyond their financial returns. Google sells Anthropic cloud infrastructure. Amazon does the same. Their interests in how Anthropic allocates compute, prices its API, and prioritizes research are not aligned with the dispassionate fiduciary oversight the LTBT was designed to provide. The scholarship calls this the "superstakeholder problem." When your biggest investors have operational relationships with you, the governance mechanism designed to check profit pressure becomes something closer to a conflicted committee.

The Harvard Law Review analysis also flagged a structural circularity: Anthropic describes the LTBT's purpose as identical to its own — responsibly developing AI for humanity's long-term benefit. That means the trust has no independent standard against which to measure the company's behavior. If Anthropic decides that a given safety trade-off serves the long-term benefit of humanity, the trust has no grounds to disagree. The mechanism checks shareholder pressure. It does not check the company's own judgment about what constitutes responsible development.

Anthropic has said almost nothing about what the S-1 actually contains. What is known: Anthropic raised $65 billion in Series H funding last week at a $965 billion post-money valuation, and its revenue run rate crossed $47 billion earlier this month, according to the company's own blog. It has also reportedly targeted its first profitable quarter in Q2 2026, with an operating profit around $559 million, per outlets citing unnamed reports — a milestone it has not confirmed publicly.

The company has spent the last two years in an active dispute with the U.S. military. The Pentagon labeled Anthropic a security threat and excluded it from defense contracts; Anthropic called the blacklisting retaliatory and filed suit. The dispute is itself a test case for what the LTBT is supposed to protect: a company that turned down military work on safety grounds, now facing regulatory retaliation that its competitors do not face. Whether the trust advocated for Anthropic in that dispute, or whether it stayed silent, is not publicly known.

The $965 billion valuation, the $47 billion revenue run rate, and the prospect of a profitable quarter suggest the market window is now. What the S-1 will not immediately reveal is whether the trust designed to keep the most powerful AI company in the world aligned with its stated mission has the independence, the information, and the membership to do that job — or whether it was built to reassure investors rather than constrain them.

The answer lives in a sealed filing. When it surfaces, the governance exhibit will show what the trust actually controls, who sits on it, and what authority it has to override the people who run Anthropic. That document is the next factual checkpoint for anyone trying to understand what, exactly, a $965 billion AI safety bet is worth.