Anthropic Is Going Public. Control Is Staying Somewhere Else.

Public capital is not public control

The filing is preliminary. Anthropic has applied to list its Class A common stock on Nasdaq under the symbol ANT, but the number of shares and price range remain blank, and the registration statement has not become effective. Those details will change. The control architecture is already described with unusual precision.

Each Class A share receives one vote. The Founder LLC, initially owned by Dario Amodei, Daniela Amodei, and Anthropic's five other co-founders, will hold one Class F share. That share has no ordinary economic rights, but it carries an aggregate 50.1% of the voting power on matters submitted to stockholders generally, including election of directors other than those reserved to Class T. The Long-Term Benefit Trust holds the Class T share, also without ordinary economic rights, and generally elects and removes the Class T directors who constitute a majority of the authorized board.

The filing states the practical result directly: the trust and Founder LLC collectively can elect all board members. Public investors may own economic exposure and a formal vote, yet they will not possess the control normally associated with a majority of voting shares. This is more than the familiar dual-class founder structure. Power is split between founders and a purpose trust, each holding stock whose governance rights are intentionally separated from economics.

That separation can be defended. The people making a safety decision may be less tempted to protect the next quarter's share price if their special voting stock has no claim on dividends or appreciation. But insulation from financial incentives is not the same as independence from personal conviction, institutional loyalty, reputation, or groupthink. A governance document can change who holds power. It cannot prove that power will be exercised well.

Anthropic is not asking public investors to discover the control structure later. It is asking them to buy into it now.

The trust has power, but no beneficiaries

The Long-Term Benefit Trust is a Delaware noncharitable purpose trust. Its amended agreement says the trustees must use their powers to ensure that Anthropic responsibly balances stockholder pecuniary interests, the interests of people materially affected by the company, and the responsible development and maintenance of advanced AI for humanity's long-term benefit. The agreement then says the trust has no other purpose and no beneficiaries.

That language is conceptually important. Traditional trust vocabulary encourages a reader to imagine an identifiable person whose interests the trustee must protect. This trust is organized around a purpose instead. The trustees are meant to be financially disinterested and to bring expertise in AI safety, national security, public policy, and social enterprise. Their principal corporate power is selecting a majority of the board, not running Anthropic day to day.

Purpose governance solves one problem and creates another. A beneficiary can often recognize when a trustee has failed that beneficiary. Humanity's long-term benefit has no single authorized interpretation. Reasonable people can disagree about whether the safer decision is to delay a model, deploy it broadly to improve defensive capability, cooperate with a government, refuse a government, publish research, or keep it secret. The more expansive the purpose, the more important the process used to interpret it.

The right accountability questions therefore concern selection, removal, conflicts, information, and explanation. Who chose each trustee? What relationships could influence judgment? What evidence reaches the trustees? When do they disagree with management? What happens when the founder bloc and the trust prefer different directors? The filing supplies legal mechanics. Public legitimacy will require a continuing record of how those mechanics operate.

Public benefit status is a balancing instruction, not an answer

Anthropic is also a Delaware public benefit corporation. Its stated corporate purpose is the responsible development and maintenance of advanced AI for humanity's long-term benefit. Delaware law requires the board to balance three things: stockholders' pecuniary interests, the best interests of people materially affected by the company's conduct, and the public benefit identified in the charter.

That is a genuine legal difference from an ordinary corporation, but it should not be romanticized. Section 365 also says that a director does not, merely because of the public benefit provisions, owe a duty to a person who has an interest in that benefit or is materially affected by the company's conduct. For a balancing decision, an informed and disinterested director receives substantial protection unless the decision is one that no person of ordinary, sound judgment would approve.

In other words, the statute gives directors room to consider safety and social consequence without reducing every dispute to immediate shareholder value. It does not tell them how to resolve the hardest tradeoffs. It does not appoint the public as a client. It does not turn a broad promise to humanity into a judicially supervised operating manual.

Lawyers should recognize this pattern beyond Anthropic. Governance language can authorize a decision and still leave the decisive judgment with people. The most careful charter cannot determine how much evidence is enough, how much risk is acceptable, or which affected community matters most. Those questions return through board materials, committee work, conflicts procedures, disclosure controls, and the quality of dissent inside the institution.

The controlled-company label has consequences

Because the trust can elect a majority of the authorized board, Anthropic qualifies as a controlled company under Nasdaq's rules. Controlled companies can claim exemptions from requirements governing a majority-independent board and independent compensation and nominating committees. The prospectus says Anthropic expects to rely on the exemption only for its nominating and corporate governance committee, which will not be entirely independent. It says the board currently has an independent majority and its audit and compensation committees are entirely independent.

That limited use matters, but so does the function involved. A nominating committee influences who will monitor management after today's founders, trustees, and directors are gone. Anthropic's architecture is expressly designed to survive leadership transitions. Independence at the point of succession is therefore not a peripheral compliance detail. It is part of the product being offered to investors and the public: durable stewardship rather than permanent reliance on current personalities.

Nasdaq's exemption is not a finding that Anthropic's governance is defective. It recognizes that a controlling holder already possesses the practical power to select directors. The prospectus appropriately warns that investors may not receive the same protections as stockholders of companies subject to every ordinary requirement. The proper response is not to pretend the exemption is a scandal. It is to evaluate the substitute protections with equal seriousness.

The structure will be tested when the mission is expensive

Good governance is easiest when mission, revenue, national policy, employee preference, and investor return all support the same decision. Anthropic's structure matters when they do not. A model may create obvious commercial opportunity and uncertain public risk. A government contract may advance national security while conflicting with internal safety judgments. A delay may protect the public and hand market share to a less cautious competitor. A disclosure may increase accountability and reveal capabilities that should not spread.

No charter eliminates those conflicts. The most a structure can do is place the decision with people who have authority, information, independence, and a process capable of producing an answer that can be examined. Anthropic has created more institutional machinery for that task than many technology companies. It has also concentrated extraordinary power in bodies that ordinary public shareholders cannot replace through ordinary voting.

That bargain may prove wise. The public filing is valuable precisely because it lets investors, regulators, employees, customers, and critics evaluate the bargain before a crisis. The next standard should be operational transparency: current trustee identities and qualifications, clear conflict disclosures, meaningful explanation of succession, and reporting that shows how the PBC purpose affects real decisions without exposing sensitive security information.

An IPO will not settle whether Anthropic is a responsible steward of advanced AI. It will make the governance claim public, durable, and testable. The company should be judged neither by the elegance of its documents nor by suspicion of anything unconventional. It should be judged by what happens when the structure has to choose, in public, between the purpose it was built to protect and the pressure it was built to resist.

Mission protection is not accountability's opposite. It becomes credible only when outsiders can see how power was used in the hard case.