Anthropic built its brand on one idea: we are the responsible AI company. Constitutional AI. Careful deployment. On Friday, April 3, 2026, the company filed paperwork with the Federal Election Commission to launch a political action committee called AnthroPAC.
PACs are not unusual in the technology sector. Google, Microsoft, and Amazon all have them. What makes AnthroPAC notable is the company filing it: Anthropic’s public identity is built specifically around AI safety, and CEO Dario Amodei has written publicly about existential risk from advanced AI. A company whose stated mission includes making sure AI is developed safely is now also a company that funds the campaigns of the people who regulate it. Whether those two things are in tension, or are simply how a regulated industry operates, is the question this piece tries to answer with public evidence rather than with anyone’s private read on the room.
What AnthroPAC Actually Is
Per AnthroPAC’s FEC Statement of Organization (Committee ID C00946111), AnthroPAC is a traditional corporate PAC, structured as a separate segregated fund connected to Anthropic PBC, funded by voluntary employee contributions capped at $5,000 per person per year. Allison Rossi, Anthropic’s treasurer, signed the filing from the company’s San Francisco headquarters; Jared Powell is listed as assistant treasurer. A bipartisan board will decide which House and Senate candidates receive money, filtered through AI policy relevance. All donations and disbursements will be reported through FEC filings, which are public.
This is different from a super PAC in a way that matters. Super PACs accept unlimited money but cannot give directly to campaigns. AnthroPAC can write checks directly to candidates but is limited to employee contributions of up to $5,000 per person per year, not corporate treasury funds.
AnthroPAC only just formed. As of this writing it has not yet disclosed any actual disbursements, so there is no FEC record yet of which candidates or parties it has funded. Separately and prior to AnthroPAC’s formation, individual people closely associated with Anthropic have well-documented personal political giving records: early investor Dustin Moskovitz has donated roughly $110 million to political causes over the years, the large majority to Democratic-aligned recipients, and Anthropic board member Reed Hastings has donated tens of millions to Democratic campaigns and committees, including a reported $7 million to a pro-Harris super PAC in 2024. Those are personal donations by individuals, made in their own capacity and largely before AnthroPAC existed. They are not AnthroPAC contributions, and this piece is not treating them as the same thing. Whether AnthroPAC’s own future disbursements turn out to be genuinely bipartisan or lean toward one party is a question only its own FEC filings, once they exist, can answer.
The Pentagon Fight That Explains the Timing
AnthroPAC arrives during a legal dispute between Anthropic and the Trump administration. Anthropic wanted contractual language requiring Claude’s use in military contexts to follow the company’s Acceptable Use Policy; the Pentagon wanted unrestricted access. In February 2026, Defense Secretary Pete Hegseth’s department designated Anthropic a “supply chain risk,” and a reported $200 million contract was cancelled. Anthropic filed suit. A federal judge in California has since blocked the Pentagon from taking further punitive action against Anthropic pending the litigation, finding the government’s response likely implicated the company’s First Amendment and due process rights. The dispute remains active.
Anthropic’s position, that an AI company should have a say in how the government deploys its models, is a stated safety principle. It is also a business and legal fight that benefits from allies in Congress. AnthroPAC exists at that intersection. That is a defensible business rationale. It is also inherently a political one, and it is fair to note that a company whose core public identity rests on safety-first positioning is now also playing the same access-buying game every other major tech company plays.
The $300 Million Context
AnthroPAC does not exist in isolation. Reporting close to AnthroPAC’s filing date put AI industry spending on the 2026 midterms above $300 million, with earlier March reporting (via The Washington Post) putting the industry total nearer $185 million, a range that reflects both a moving target and differences in what different outlets counted. Leading the Future, a super PAC backed by OpenAI’s Greg Brockman and Andreessen Horowitz, has raised over $100 million. Separately from AnthroPAC, Anthropic itself donated $20 million to Public First Action, a bipartisan advocacy group focused on AI safeguards, in February 2026, before AnthroPAC’s formation.
The committees that matter here are Senate Commerce and House Energy and Commerce, which are drafting liability frameworks, chip export controls, copyright rules for training data, and immigration policy for AI talent. The regulatory backdrop is real: dozens of AI-related bills are active across state legislatures, Tennessee has restricted AI systems from representing themselves as mental health professionals, New York’s RAISE Act targets frontier models above a compute threshold, and California’s SB 53 requires safety documentation and whistleblower protections. For a company training frontier models, this is the environment AnthroPAC is meant to operate in.
What Remains an Open Question
The tension between Anthropic’s safety-focused public identity and its new role as a political spender is real, and it does not require appeal to private conversations to establish: it follows directly from the public facts above. A company that argues AI poses serious enough risks to warrant careful, sometimes restrictive deployment decisions, including refusing a Pentagon contract over guardrails, is also now funding the campaigns of the legislators who will decide how much deference to give companies making exactly those calls. That is not necessarily hypocrisy. Lobbying and safety advocacy are not mutually exclusive, and every other major AI lab is doing some version of the same thing. But it is a real tension, and it is one Anthropic will have to answer for in public, through its actual FEC disbursement records, not through anything this piece can settle in advance.
Where This Goes
The clearest test is straightforward and entirely public: watch AnthroPAC’s FEC filings as disbursements start. If the giving turns out to be genuinely split across parties, the bipartisan framing holds up. If it consistently favors one party, that will be visible in the public record within a few reporting cycles, and the credibility cost of having claimed bipartisanship will be immediate.
For Anthropic specifically, AnthroPAC is one piece of a broader expansion: the company is acquiring biotech startups for $400 million, restructuring its pricing model, fighting the Pentagon in court, and reportedly preparing for a possible IPO. Whether a company can simultaneously build frontier AI, lobby for favorable regulatory treatment, and retain credibility as a voice on the risks that regulation is meant to address is a genuinely open question, one that will be answered by what AnthroPAC’s own public filings show over the coming election cycles, not by how the story is framed on day one.
Sources: AnthroPAC FEC Statement of Organization, Committee ID C00946111 (fec.gov); reporting on the April 3, 2026 filing from Decrypt, the Washington Examiner, and Implicator.ai; reporting on the Pentagon dispute and federal court ruling. Updated 2026-08-18: this article previously relied on private, off-record, and NDA-sourced claims about reactions inside the AI safety community, which have been removed in line with this site’s rule that private context does not become public substantiation. It also previously applied specific partisan-giving percentages (82%, 99.8%) to Anthropic and AnthroPAC that actually describe personal donations by individuals affiliated with the company, not the company or the PAC itself; those figures have been removed and the individual donations are now described separately and attributed to the individuals who made them, not to Anthropic or AnthroPAC.