A 501(c)(4) is a category of US tax-exempt nonprofit, named for the section of the federal tax code that creates it, built for “social welfare” work rather than charity. Unlike the better-known 501(c)(3) — the status behind most charities and foundations — a 501(c)(4) is allowed to spend money on politics, including ads that name specific candidates, as long as that isn’t its main activity, and it generally doesn’t have to disclose who funds it. That mix of political reach and donor secrecy has made 501(c)(4)s the vehicle of choice for one of the bigger fights in Washington right now: how far the US government should go in regulating artificial intelligence. AI companies are now funding both sides of that fight, and the money moves mostly through structures like this one.
What sets a 501(c)(4) apart from a charity
The US tax code sorts nonprofits by what they’re allowed to do. A 501(c)(3) — the status most charities and foundations use — is barred from political campaigning altogether, and donations to it are tax-deductible. A 501(c)(4) trades that deduction away for more freedom: it can run political ads and back policy fights, as long as that stays secondary to a broader “social welfare” mission and under roughly half of its total spending. Crucially, a 501(c)(4) generally doesn’t have to publish its donor list — which is why the structure is sometimes called “dark money”: the spending is public, the people paying for it often aren’t.
How it differs from a super PAC
A super PAC sits at the other end of the same toolkit. It can also raise unlimited money, but it exists specifically to support or oppose candidates, and by law it must publicly disclose its donors. In practice, real advocacy operations often run both at once: a 501(c)(4) handles “public education” messaging with largely anonymous funding, while one or more affiliated PACs handle the direct campaign spending with disclosed contributions.
Why AI companies are suddenly major players
That exact pairing describes the two operations now competing over AI regulation in the US. Leading the Future launched in 2025 as a super-PAC network backed by more than $100 million from figures including OpenAI president Greg Brockman, Andreessen Horowitz, and Palantir co-founder Joe Lonsdale; it argues against strict state-level AI safety laws in favor of one lighter national standard. On the other side, Public First Action — a 501(c)(4) founded by former US representatives Brad Carson and Chris Stewart — has raised roughly $80 million across its network, with Anthropic as its single largest donor, arguing instead for federal AI safeguards that don’t override tougher state rules. Public First Action’s own money is restricted to “education and policy” work rather than election ads; direct campaign spending runs through affiliated committees such as Jobs and Democracy PAC and Defending Our Values PAC.
Why it matters
None of this is illegal or unique to AI — 501(c)(4)s and super PACs fund political advocacy across every major US industry, from energy to pharmaceuticals. What’s new is the scale AI companies are now bringing to it: well over $100 million combined, deployed by firms that were, until recently, mostly research labs. For anyone trying to follow why state AI safety bills keep stalling or advancing, or why Congress keeps debating whether to override state AI law altogether, the money behind that debate runs largely through structures like these. It’s also a preview of how AI governance is likely to get decided in practice — not only by regulators writing rules, but by which side outspends the other before the rules are written. Understanding the 501(c)(4)/super-PAC split explains how a company can accurately say its donation “can’t fund a candidate” while the broader network it bankrolls still shapes elections.
In the news
Anthropic’s donation is a live example of this pattern: it recently doubled its commitment to Public First Action to $40 million, citing the pace of AI capability gains as its reason for wanting stronger federal safeguards in place.