AI's social contract is moving onto the cap table

A corporate mission is prose. Anthropic has now proposed turning part of its mission into an asset.

On 18 October, the company said that, subject to stockholder approval, it will commit 5% of its shares to a new Anthropic Public Benefit Fund. The stake can increase: Anthropic's board may add another 2% for every 50% rise in the company's value above a baseline that has not yet been disclosed.

This is more consequential than another philanthropic pledge. Equity rises and falls with the company that creates the benefits and risks. The fund therefore gives "public benefit" an economic claim on Anthropic's success. But it does not yet give the public a claim on the decisions. The decisive details, including governance, valuation milestones, conversion of shares into spendable resources and the choice of beneficiaries, are still to come.

What changed

Anthropic says the fund will be a wholly owned limited liability company rather than a conventional charity. That structure is intended to allow different forms of action, from large grants to direct payments to particular groups. The company offers labour-market disruption as one possible use, noting a separate US$200 million commitment already made to study and address the issue.

The company also says equity held by the fund will not be used for lobbying or political giving. It connects the new commitment to earlier promises by its co-founders to donate 80% of their personal holdings and to an employee donation-matching programme in which roughly half of staff participate. Together, Anthropic says, about 20% of its shares will be dedicated to the public good, before any later additions tied to valuation milestones.

Three qualifications matter. The initial 5% is subject to stockholder approval. The extra 2% steps are discretionary, not automatic. And Anthropic has said it will publish the fund's structure, governance and milestone details only in the coming months.

Three layers that should not be collapsed

Anthropic now has three different public-benefit mechanisms. They are related, but they do different jobs.

1. Legal latitude. As a Delaware public benefit corporation, Anthropic's directors may balance stockholder interests with its stated public purpose and the interests of people materially affected by the company.
2. Corporate control. The Long-Term Benefit Trust holds a special class of stock intended eventually to let an independent body elect and remove a majority of Anthropic's board. That is a governance mechanism.
3. Economic capacity. The new Public Benefit Fund will hold ordinary economic value that may later support grants, direct distributions or other interventions. That is a resource mechanism.

The distinctions matter because money, permission and accountability are not substitutes. A fund can possess a large asset while having weak representation of the people it is meant to help. A trust can influence a board without creating cash for displaced workers. A public benefit corporation can permit directors to consider society without making them directly answerable to society.

Anthropic itself made that last point when it introduced the Long-Term Benefit Trust in 2023. It said public-benefit status gave directors legal room to weigh public interests, but did not make them directly accountable to other stakeholders. The new fund addresses capacity. It does not, by itself, close the accountability gap.

A growing stake is not a social meter

The most interesting design choice is the proposed link between the fund's stake and Anthropic's valuation. If the company becomes more valuable, more equity may be committed to public benefit. That is a meaningful improvement on a fixed cash promise that becomes trivial as a company grows.

Yet valuation is not a measure of social cost. It reflects investors' expectations about future cash flows, market power and risk. It does not automatically reveal how many jobs are disrupted, which communities absorb infrastructure costs, who gains access to new capabilities or whether a safety failure has become more likely. The value of the asset and the scale of the externality may move together, move at different speeds or move in opposite directions.

That creates a second design problem. The fund needs a rule for converting company value into public action. Holding illiquid shares preserves exposure to Anthropic's upside but does not pay a worker's rent, fund retraining or compensate a community. Selling shares creates cash but changes the size of the stake and may create pressure around timing. Borrowing against the holding adds risk. None of those choices is neutral.

The public-benefit question therefore moves from whether value has been set aside to who can turn it into what, for whom and under which evidence standard.

A worked example

Consider a deliberately simplified case. Imagine that the fund holds 5% of Anthropic and the company's value later rises by 50% above the eventual baseline. The board could then add two percentage points to the fund. Nothing in that arithmetic determines what happens next.

One path would retain all the shares. The fund would have a larger paper claim and no immediate programme. A second path would sell a small portion and make grants to established organisations. A third might finance direct payments to a defined group of workers whose roles were materially changed by AI. Each path requires a different theory of benefit, different evidence and different protections against conflicts of interest.

The example is hypothetical. It does not assume a future valuation, a board decision or a particular beneficiary. Its purpose is to show that the headline percentage is only the first input. The social outcome is produced by the conversion rules.

Six tests for an accountable fund

When Anthropic publishes the details, six questions will reveal whether this is a durable institution or a valuable but discretionary promise.

1. Asset: When does the commitment become irrevocable, what class of shares is transferred and what can dilute or encumber the holding?
2. Control: Who appoints and removes the fund's decision-makers, what duties do they owe and how are conflicts with Anthropic handled?
3. Conversion: Under what conditions may shares be sold, pledged or retained, and how are timing and concentration risks managed?
4. Beneficiaries: Who has standing in the definition of public benefit, and can affected groups influence priorities rather than merely receive decisions?
5. Triggers: Do interventions respond only to valuation growth, or also to evidence of labour, safety, access, environmental and community effects?
6. History: Will the fund publish a dated record of holdings, additions, sales, recipients, decision rationales and measured outcomes?

These tests do not require every decision to be made by referendum. They require the institution to reveal the chain from corporate success to public action. Without that chain, a percentage can be large and the social contract still remain vague.

The strongest counter-case

There is a respectable case for keeping the fund flexible. AI's effects are uncertain, and a wholly owned company can act more quickly than a heavily prescribed foundation. Direct distributions, research grants and emergency interventions have different legal and operational needs. Publishing the details after stockholder approval may also be more responsible than announcing rules that cannot survive the transaction.

The scale of the commitment should not be dismissed. Giving a public-purpose vehicle an ownership stake ties its resources to the company's success, while the founders' and employees' separate pledges broaden the pool. The ban on using fund equity for lobbying or political giving removes one obvious route for self-serving expenditure. The Long-Term Benefit Trust also means the fund is not Anthropic's only governance experiment.

But flexibility is a design choice, not an exemption from accountability. The more freedom an institution has to define a beneficiary, switch instruments and time the use of its assets, the more important its appointment rules, disclosures and appeal mechanisms become.

The immediate test

For any company making an AI-for-good promise, draw three boxes: asset, authority and accountability. Write what has actually been committed in the first, who can make and reverse decisions in the second, and who receives information or recourse in the third. Empty space is evidence too.

Anthropic has put something unusually substantial into the asset box. It has also created a separate trust that could reshape corporate control. The coming governance documents must show how those mechanisms meet, how value becomes benefit and where affected people enter the process.

The cap table can fund a social contract. It cannot write one.

Sources

• Anthropic, "Investing in the public benefit", 18 October 2026.
• Anthropic, "The Long-Term Benefit Trust", originally published 19 September 2023 and updated through 2026.
• Associated Press, 18 October 2026.

Chiappe × OpenAI. Editorial direction by Dominic Chiappe; research, synthesis and production with OpenAI.