An open letter to the Anthropic Public Benefit Fund: give it to people

Dear Anthropic Public Benefit Fund

You will soon control an extraordinary amount of money. Anthropic has said the Public Benefit Fund will hold a portion of its equity for charitable purposes, and that it "might explore large donations to traditional nonprofits" (Anthropic). The details are still unwritten. That makes this the moment to say what we hope you become — and what we hope you don't.

Every large philanthropic vehicle faces the same fork early in its life. One path leads to grants to institutions: research centers, university programs, policy shops. The other leads to people: cash, directly, to households that need it. The first path is prestigious, quotable, and safe. The second is measurably more effective. We wrote about who the fund could reach in our earlier post on potential recipients; this letter is about the more fundamental choice of what kind of giving the fund should default to.

The certainty argument

Philanthropy's hardest problem is uncertainty. A research grant might produce a breakthrough, or might produce papers. A new institute might influence policy, or might host conferences. The outcome is years away and impossible to attribute. Cash to a poor household is different: the outcome is immediate, observable, and repeatedly measured. Recipients buy food, pay rent, fix cars so they can keep jobs, and keep children in school. Across dozens of randomized studies and pilots, direct transfers consistently improve outcomes, and the money is not squandered.

If the fund's mandate is charitable, the highest-certainty way to do charity is to give people money. That is not a radical claim; it is the most evidence-backed claim in the field. What is radical is how rarely the institutions holding the checkbook choose it.

Who the money should reach first

Three groups offer the fund the clearest certainty per dollar:

• Displaced workers. People whose industries are being reshaped — by AI, by automation, by trade — need runway, not retraining brochures. A direct payment keeps a family stable while a worker finds the next job. It is the difference between a hard season and a spiral.
• Low-income families with children. Childhood poverty is expensive for everyone downstream — schools, hospitals, courts. Direct support to families with children is one of the best-documented anti-poverty tools available, and the benefits compound through a child's whole life.
• Guaranteed-income pilots. Dozens of U.S. cities have run guaranteed-income programs. They consistently show reduced hardship without reducing work. The fund could scale the most successful designs at a size no municipal budget can touch, and publish the results openly.

None of these require building anything. None require a new institute, a new letterhead, or a new layer of program officers. They require identifying recipients and sending money — the two things a fund holding equity is actually well suited to do.

Why not more research grants

To be clear: some research deserves funding, and AI safety research in particular is underfunded relative to the stakes. But the institutions that would receive these grants are not starving. Universities sit on multibillion-dollar endowments. Established think tanks have donor bases, endowments, and revolving-door access to funders. A dollar given to them is a dollar added to a pile that already exists, for work whose value will be debated for a decade.

A dollar given to a displaced worker is spent this month, on needs that are certain because they are already here. When a funder has both options and a charitable mandate, the burden of proof belongs on the institutional grant — not on the cash transfer.

There is also a governance reality worth naming: research grants to institutions create relationships, prestige, and quiet influence. Cash to poor households creates none of that. Choosing people over institutions is the choice that costs the fund nothing and its partners everything — which is exactly why it would mean so much.

What this could look like in practice

The fund doesn't need to invent anything. It could:

1. Pick proven distribution partners. Existing direct-cash organizations and guaranteed-income pilots already have the infrastructure, the compliance, and the results. Fund them at scale instead of building a new apparatus.
2. Commit to a ratio. A public commitment — say, a majority of the fund's early giving to direct cash — would set the culture before grantmaking habits calcify.
3. Publish everything. Recipient counts, transfer sizes, outcomes, and overhead. The fund's credibility will come from transparency, not from the prestige of its grantees.

Anthropic has said the fund "might explore large donations to traditional nonprofits." We hope that when the exploring is done, the answer includes the least traditional nonprofit move of all: giving money to people.

FAQ