Anthropic’s donations can’t be used to influence elections — despite what everyone thought
The company's money isn’t allowed to be used in the midterm battles. Without it, pro-safety candidates may be even more outgunned than expected
Earlier this year, Anthropic donated $20m to Public First Action — a donation which was, at the time, widely expected to be used to fund political ads for members of Congress who support more stringent AI safeguards.
But that is not the case, Transformer has learned. “Anthropic restricted its donation from being used to influence federal elections,” an Anthropic spokesperson told Transformer.
“Anthropic’s donation to Public First Action was a contribution to a 501(c)(4) exclusively in support of its mission to educate the public on AI policy and promote safe and responsible AI.”
The revelation raises questions about who is funding the $3.48m that Public First Action’s associated super PACs have spent on elections to date.
It also calls into question whether, without Anthropic’s backing for electoral spending, advocates for stronger AI safeguards have as much sway in the midterm elections as was previously believed.
As a 501(c)(4) nonprofit, Public First Action must comply with IRS restrictions, which prohibit “political activity on behalf of or in opposition to candidates” — such as running ads supporting a candidate in an election — as the organization’s “primary” activity. Typically this has been interpreted as prohibiting political activity from making up 50% or more of the organization’s spending. It is possible, however, that Public First Action will in fact be forced to spend much less than that.
Two months ago, Public First Action leader Brad Carson told the New York Times his group had raised nearly $50 million. But in a statement to Transformer last week, spokesperson Anthony Rivera-Rodriguez did not answer a question about whether Carson was referencing money the 501(c)(4) had raised or the PACs, provide a current fundraising total for any of the groups, or reveal what portion of Public First Action’s contributions could be used for election spending.
The three super PACs aligned with Public First Action — Public First, Defending Our Values, and Jobs and Democracy — have to date disclosed $3.48m in spending, but very little about their funding. FEC filings currently only reveal a singular $50,000 contribution from Public First Action to the Republican super PAC Defending Our Values. More details are expected to be released in new quarterly filings this week.
Neither Anthropic nor Public First Action has previously disclosed the restriction on Anthropic’s donation, despite a string of prominent news articles indicating that Anthropic’s money would fund the group’s super PACs. Articles in Bloomberg, the Washington Post, and many more outlets, including Transformer, reflected the assumption that Anthropic was, via Public First Action, funding super PACs and political activity. A company blog announcing Anthropic’s donation was vague about the topic of direct campaign influence.
The ambiguity likely helped Public First. The perception that Public First was backed by Anthropic made it look like a credible counterweight to Leading the Future, which has raised a more than $50m war chest to support candidates backing weaker regulation and which has ties to that company’s biggest competitor, OpenAI. Given Anthropic’s extraordinary growth, the reports led to an assumption that many more millions could flow in later. Anthropic’s gift also drew attention to Public First Action, potentially attracting more donations from those who hold similar values around AI safety and concerns about existential risk, and giving politicians the cover to push for stronger regulations without fear it will cost them an election. By not correcting the record the perception of electoral firepower persisted — without actually having to deploy it.
Anthropic only disclosed the restrictions after Transformer asked whether the donation complied with campaign finance laws that prohibit government contractors from making political contributions.
The line between “influencing federal elections” and “educating the public on AI policy” is blurrier than it may seem. Public First Action recently ran an advertising campaign saying that “New Jersey needs leaders like Congressman Josh Gottheimer,” and telling voters to call him and urge him to “stand strong for AI safeguards.” That sort of ad does not necessarily fall under the IRS definition of “political activity” because it is not directly telling viewers how to vote.
But not being able to directly fund ads supporting or opposing candidates severely restricts what Anthropic’s money can be used for — and in turn whether Public First is able to go toe-to-toe with its extraordinarily well-funded opposition.
Leading the Future and its affiliated super PACs have disclosed $50m in donations to date, and claim to have raised $125m from donors including OpenAI president Greg Brockman and his wife Anna, and Marc Andreessen and Ben Horowitz of venture capital firm Andreessen Horowitz. A partisan dark money group led by Trump’s former deputy chief of staff Taylor Budowich also said it plans to spend $100m (though as a 501(c)(4), the majority of that also cannot be used for political campaigns.)
Public First’s Carson has always conceded that Public First Action has a fraction of Leading the Future’s war chest, asserting that it doesn’t need as much money because public opinion is on his side. Indeed, polls consistently show that Americans of all stripes demand more AI safeguards. But Anthropic’s admission suggests the group may be even more outgunned than was previously believed: less a competitive counterweight than a paper tiger.