Dario Amodei made a bet in July 2025 when Anthropic signed a $200 million Pentagon contract: a private AI company could set use-case limits on its own models and win in court. That bet paid off on March 26, when U.S. District Judge Rita Lin in San Francisco issued a preliminary injunction blocking the Trump administration's designation of Anthropic as a "supply chain risk," declaring the government's conduct was, in her words, "classic First Amendment retaliation." DOJ filed its appeal six days later. And on April 3, Anthropic filed paperwork to create AnthroPAC, turning a legal victory into something more enduring: a political operation.
The move is escalation dressed as disclosure. AnthroPAC, registered with the FEC as a Lobbyist/Registrant PAC with Allison Rossi named as treasurer, is structured as an employee-funded vehicle with voluntary contributions capped at $5,000 per person annually, according to Axios. It will raise modest sums against the $125 million-plus that OpenAI has reportedly bundled into its own super PAC network. But the PAC is not the story. The real money moved in February, when Anthropic donated $20 million to Public First Action, a 501(c)(4) that funded the Jobs and Democracy PAC. That vehicle spent $1.6 million supporting Representative Valerie Foushee in her North Carolina Democratic primary, part of a broader effort to cultivate a bipartisan coalition for what Anthropic describes as "sensible AI regulation."
The $20 million donation is the actual lever. It is also opaque by design: 501(c)(4) organizations are not required to disclose their donors publicly, which means the flow of AI-industry money into electoral politics is obscured at precisely the point where disclosure matters most. The counterargument is obvious: it is also a political investment designed to limit OpenAI's influence as both companies prepare for IPO.
Anthropic has been building toward this moment methodically. It registered its first in-house lobbyist in 2024, then retained Continental Strategy in April 2025 and Fierce Government Relations by November. By December it had three more firms on retainer. Total lobbying spend for 2025 was $3.13 million, quadruple the prior year. Sarah Heck now runs policy at the company, succeeding co-founder Jack Clark, who departed to focus on the Anthropic Institute. The departure was not cosmetic: Clark built the company's initial Washington posture around principles and publications. Heck is running it like a political operation.
The origin of the current dispute traces to contract negotiations last September, when the Pentagon sought "all lawful uses" language for Claude's deployment on classified networks. Anthropic refused. It drew two explicit red lines: no Claude for fully autonomous weapons, and no Claude for domestic mass surveillance of Americans. When those negotiations collapsed, the administration moved to blacklist the company entirely. The Pentagon labeled Anthropic a "supply chain risk" on February 27, a designation never before applied to an American company.
Judge Lin's ruling was a significant vindication of that stance, and of Amodei's willingness to litigate rather than comply. The preliminary injunction blocked the supply-chain designation and the executive directive to phase out Claude across federal agencies. The DOJ appeal keeps the question alive, and the Ninth Circuit's resolution will shape whether AI companies have a judicially enforceable right to restrict how their models are used, or whether those restrictions can be treated as government procurement leverage.
AnthroPAC's filing lands while that appeal is pending. It is both a hedge and a next step. If the litigation succeeds, the PAC gives Anthropic a vehicle to fund candidates who support its regulatory vision. If it fails, the company will have already built political infrastructure independent of any single court ruling. The $5,000-per-person cap means the PAC itself will not rival OpenAI's spending. What matters is the template it represents: an AI lab using political spending to lock safety constraints into regulatory reality, rather than simply hoping courts or Congress will preserve them.
The 2026 midterms will be the first real test of which model wins. Amodei's bet is that the answer is not decided in a courtroom alone.
