FUNDING
How the Institute Is Funded
The Recursion Institute takes no money from the AI companies it evaluates. That is not a courtesy or a preference — it is the product. An institute that documents how these systems fail, and asks their makers to fix it, has exactly one asset that gives its findings weight: it answers to no one it studies. Everything below is how that independence is kept honest, stated plainly enough to check.
The Institute is a for-profit LLC, founded in Florida, funded by the people it works for — readers, and the institutions that want independent user-side AI-safety documentation to exist. It is not a charity, and it does not want to be one. The nonprofit model would trade the very independence the work rests on for a tax status; the model that fits the mission is the one Consumer Reports and reader-funded newsrooms like 404 Media proved — accountable to your audience, beholden to no subject, and free to publish whatever the evidence says.
The one-line version. Audience money first; clean, no-strings grants as add-ons; never a dollar from the labs the Institute documents. Funding is ranked by distance from the companies under evaluation — not by size — because the distance is the credibility.
Where the money comes from
Every source below shares one property: none of it can buy a finding, and none of it comes from an entity the Institute rates. That is the whole test a funding source has to pass.
- Readers. Voluntary contributions — one-time or recurring — from people who want this work to exist and to stay unaffiliated: the core engine. The people who support it are the people it protects, which is the cleanest funding relationship there is: it aligns the Institute with its audience and no one else. Everything the Institute publishes stays free; support is optional. Support the Institute →
- Education. Paid courses and practitioner training for the professions that meet AI harm firsthand — lawyers, clinicians, educators, journalists. The people learning the method are not the systems being evaluated, so there is no conflict to manage.
- Syndication. Licensing the Institute’s reporting and research to larger outlets, with the Institute keeping ownership. Money from publishers who want to carry the work — never from its subjects.
- Expert-witness & advisory work. Litigation consulting and regulator briefings drawing on the documented record. This one strengthens the independence rather than straining it: the work is aligned against developer harms, and the Institute will not consult for any AI developer it evaluates. See Services & Consulting.
- Non-dilutive grants that carry no editorial control. Grants from funders who take no equity, no board seat, and no say over what the research concludes. A grant that would require any of those is declined — the money is never worth the string. Where a grant funds a specific piece of work, that is disclosed on the work itself.
Where it does not come from
A bright line is only useful if it is named out loud. These are the sources the Institute refuses, and will keep refusing regardless of the amount:
- No money from the AI developers the Institute evaluates. Not as a grant, not as sponsorship, not as a consulting fee, not laundered through an intermediary. This is the founding rule, and it is not situational.
- No equity, ever. The Institute takes no investment that trades ownership or control for capital. A watchdog with an equity holder has a second master; there isn’t one here.
- No funder editorial say. No contributor, supporter, grantmaker, or client — at any amount — gets to shape, soften, place, or suppress a finding. Support buys the work’s continuation, not its conclusions.
- A concrete example: the AI Safety Fund. The Frontier Model Forum’s AI Safety Fund is seeded and primarily financed by the very labs the Institute documents — among them the makers of the systems in the record. It is precisely the kind of money the Institute exists to be independent of, so it is declined on positioning alone, separate from any question of eligibility. Naming it here is the point: this is what “no lab money” looks like when a real check is on the table.
The guardrails
Because the Institute is a business and not a charity, a few things follow — and they matter enough to state on their own line rather than bury in fine print.
- Not tax-deductible. Contributions are commercial support for an independent publication, not charitable donations. There is no deduction, and the Institute makes no charitable-donation claim, because it is not a charity and does not solicit as one.
- Taxable business income. Every contribution is gross income to the LLC — support for the work, not a gift to a cause. That is the honest characterization, and it is also part of why the independence holds: this is a transaction between a publication and the people it serves, on ordinary commercial terms.
- Commercial support, framed as such. The Institute asks you to fund independent AI testing and sustain this publication — never to give to a charitable cause. That framing is deliberate: it keeps the relationship clear, and it keeps the Institute plainly on the commercial side of the line, where a reader-funded for-profit belongs.
The independence pledge, in plain English. The Institute takes no money from any AI company it evaluates. No funder — reader, supporter, grantmaker, or client — receives editorial control, findings, placement, softening, or silence in exchange for support. Published work carries its falsification criteria regardless of who funded the month it was written in. If those terms are the reason you support the Institute, you have understood the model exactly.
This page is the trust asset the whole funding model rests on, so it is meant to be checked, not taken on faith. If any of it reads as inconsistent with what the Institute publishes or how it asks for support, that is a discrepancy worth raising: [email protected].
Ready to sustain the work? Support the Institute. Want to engage the applied practice that helps fund the public one? Services & Consulting.