AI safety’s ‘hard money’ may be its secret weapon in the midterms

Anthropic employees in particular are giving directly to political campaigns at an unusual clip

Credit: Getty/erhui1979

While coverage of AI money flowing into the mid-terms has mainly focused on the huge sums going into super PACs such as Leading the Future and Public First, individual donations direct to campaigns from employees of AI companies and organisations have largely flown under the radar.

That’s despite the unusually large scale of these individual donations, and their ability to directly boost the campaigns of their chosen recipients in a way the super PAC money can’t. Such ‘hard money’ donations could be AI safety’s secret weapon in the 2026 midterms.

The most significant inflows come from Anthropic employees who appear to be backing candidates such as Alex Bores and Scott Wiener that favor stricter AI guardrails than their competition. According to a Transformer analysis of FEC data, Anthropic employees are listed as the source of 302 donations totaling more than $880,000 by the end of the first quarter of this year, an average of more than $12,500 per employee donating.

Employees of OpenAI have also been donating, but at a smaller scale. In total, they are listed in filings totalling more than $300,000 across 162 donations. Many of the largest donors appear to work in areas related to AI safety.

The data is based on what is filed with the FEC, and occasionally may reflect errors in the filings. For instance, one individual was listed as working at OpenAI on a total of $17,500 in donations, despite having left for Anthropic in 2024. The FEC also only lists individual donations once they hit a $200 aggregate threshold, meaning totals for those making smaller donations may slightly undercount.

In a statement, an Anthropic spokesperson said: “Anthropic employees make their own political donations as private citizens. Like many Americans, our employees care deeply about how AI is governed, and how they choose to participate in the democratic process is up to them. Anthropic has long been public about supporting thoughtful, well-crafted policy that allows society to capture the benefits of AI while managing its risks.”

OpenAI did not respond to a request for comment.

In absolute terms, the amounts spent are far smaller than the $25m AI-related super PACs have directed towards races involving candidates friendly or unfriendly to their vision of AI regulation. But that spending comes with constraints and overheads that may limit impact. A significant portion goes to PAC employees’ high salaries, for example, while the rest can only be spent on ads and advocacy produced without coordination with the campaigns they are backing. In political circles, this has earned super PAC spending the name “soft” money.

In contrast, “hard” money, such as the donations given by many Anthropic and OpenAI employees, is given directly to campaigns, and used more strategically by candidates and their staff on precisely what they think is most effective, such as hiring staff, producing rallies, and advertising the policy messages they know to be most impactful. Frontier lab employees are providing “hard money” at levels that could help decide Congressional races.

“The campaign is the one that is able to put out the message it controls and pay the people and pay the staff, and be able to do the work on the ground,” says Brendan Steinhauser, CEO of advocacy group The Alliance for SecureAI, and a longtime political strategist who has worked for campaigns and super PACs. “There’s a benefit to having that money in house, so the candidate can direct and control it.”

The prevalence and scale of contributions from Anthropic and OpenAI employees may in part be driven by the influence of the effective altruist movement, which prioritizes putting effort, and cash, where its adherents believe it will make the most difference, based on what they see as rational assessments of its likely impact. Many in the movement, which is particularly prominent in the San Francisco Bay Area, have been attracted to building and researching AI in recent years, driven by a belief that the technology is among the most consequential and potentially risky developments in modern history.

---

How to buy an AI ‘grassroots’ movement

---

AI company employees have also found themselves with extremely high pay packages, often in the high hundreds of thousands, and occasionally multiple millions, of dollars. This gives them a significant amount of free capital, much of which they have pledged to donate. On the rationalist forum LessWrong, countless posts talk about political giving: the tax implications of donating stock, supporting policy experimentation, and advancing AI safety advocacy.

Several public threads urge support for specific candidates to which the data shows Anthropic employees are giving heavily, such as Alex Bores, who has received $186,000, and Scott Wiener, who has received more than $110,000. Bores is best known as the primary author of New York’s RAISE Act, while Wiener led efforts to pass California’s SB53, and these two state level AI bills have become de-facto models for promoting transparent AI development in several other states.

These two candidates are the top recipients of Anthropic employee money, and, notably, are supported by the Anthropic-backed AI safety advocacy group Public First, while Bores has additional support from a super PAC created by Anthropic employee Daniel Ziegler. They’ve also attracted hard money donations from a slew of other notable people tied to the Bay Area’s interlinked effective altruist and AI safety research communities, including employees at Coefficient Giving, Redwood Research, 80,000 Hours, the Alignment Research Center and others. (Disclosure: Coefficient Giving is Transformer’s main donor.)

---

How Silicon Valley sold Washington an AI race

---

There are other reasons that hard money has benefits over its soft counterpart, not least, Steinhauser says, because it may attract less negative attention. “You want to see outside groups that spend money to support you, but if they do something really dumb, or they make a mistake … it can backfire, because you don’t have control over it.”

Support from the AI accelerationist super PAC Leading the Future, for example, has become more difficult for some candidates to accept due to voter concerns over corporate influence on politics and AI policy. Leading the Future’s millions in spending against Bores has also encouraged other wealthy interests to counter by putting their own millions behind the candidate. Support from Public First, meanwhile, has led Bores’ opponents to accuse him of being beholden to billionaire interests.

That’s why Steinhauser says candidates typically want their campaigns fully funded with hard money first, and supportive PAC spending second. “I believe in the full spectrum approach,” he says. “You want your campaign fully funded, and then you’re like, I’d love to see two or three or four million [from a PAC] supporting my congressional race, but I just hope [the PAC delivers] on message.”

Hard money, of course, comes with its own set of limitations, as individuals can only contribute $3,500 per campaign, and up to $7,000 per candidate should they choose to support them in both the primaries and the general election. But many Anthropic employees are maxing out these contribution limits, which, together with other employees giving similarly sized gifts, can amount to significant sums. The median total amount given by Anthropic employees directly to campaigns is $6,500, while the median individual gift is $3,500.

Senior workers at Anthropic have given large sums across multiple campaigns, with the three biggest spenders racking up totals of more than $146,000, $87,000, and $52,000, way ahead of any individuals at OpenAI. Notably, Jan Leike, former OpenAI and Google DeepMind alignment researcher, now head of the Alignment Science team at Anthropic, has given more than $24,000. Leike left OpenAI after expressing concern the company was more interested in “shiny products” than safety, and is considered highly influential within the effective altruism movement. He has also donated to Public First’s Republican and Democratic super PACs and publicly criticized Leading the Future.

Of course, it is not uncommon for individual tech employees to organize around political efforts. Workers at Google and SpaceX gave millions of dollars in hard money during the 2024 general election, for example. However, these companies have significantly more employees, and their gifts were not as narrowly targeted on issues so central to their own industry’s interests.

OpenAI does not have a corporate PAC, but Anthropic announced the creation of its own, AnthroPAC, in April. The PAC has raised an additional $119,019 from employees, though it has not yet picked which candidates to support or oppose. It’s expected to give to both Democratic and Republican candidates.

Inevitably, the super PACs and their multi-million dollar ad spends will continue to grab attention, and for good reason. But the most durable political infrastructure is likely the groundswell of rank-and-file employee donations that go directly into campaigns without building as much distaste among voters. Anthropic employees have built that influence quietly, and may be all the more effective by doing so.

---

Correction 05/29/2026: This article was amended to remove a reference to a LessWrong post saying giving to campaigns could avoid tax. The post actually refers to giving stock to 501(c)s, including advocacy organisations such as 501(c)(4)s, not incurring capital gains tax compared to selling the stock, but this does not apply to direct campaign donations.