The $40 Million Policy Gambit
Anthropic is pouring another $20 million into Public First Action. That brings their total support to $40 million for this political cycle. It is a massive sum of money for an AI safety group, especially one that claims to be nonpartisan. The startup wants us to believe this is pure public education. But check the timing. The money flows just as Washington is debating how, or even if, to regulate frontier models.
This isn't typical candidate electioneering. Anthropic is legally barred from routing this money directly to campaigns. Instead, they are shaping the narrative. Public First Action, co-founded by former Congressmen Brad Carson and Chris Stewart, is running the plays. They are targeting candidates who favor strict AI safeguards and chip export controls. It's a calculated move. Startups don't drop $40 million on civics lessons without expecting a return on their investment.
Warring Factions: Public First Action vs. Leading the Future
The political battlefield in Silicon Valley is now split down the middle. On one side stands Public First Action, backed by Anthropic's cash and co-led by former bipartisan representatives. On the other side is "Leading the Future," a Super PAC with a massive war chest of over $75 million—some sources say up to $125 million. Their donors are the tech old guard: Marc Andreessen, Ben Horowitz, OpenAI President Greg Brockman (contributing in a personal capacity), Joe Lonsdale, Ron Conway, and Perplexity.
These two groups do not just disagree; they are in an ideological proxy war. Leading the Future wants the White House to maintain low barriers to development. They argue that heavy rules kill innovation. Public First Action, meanwhile, reported raising $80 million by June 2026. They are using single-party Super PACs like the Republican "Defending Our Values PAC" and the Democratic "Jobs and Democracy PAC" to run targeted issue ads. It is a dual-track strategy. They buy influence in both parties to block the anti-regulation lobbying from taking over.
The Digital Markets Act and EU Antitrust Enforcement Double & Triple Jeopardy
US tech giants are playing a dangerous domestic game, but they are forgetting the global board. Anthropic and OpenAI might lock horns in Washington over safety audits and FAA-style regulators. Yet across the Atlantic, the regulatory net is far more brutal. Look at how European regulators are squeezing American hardware and model partnerships. It is a reality structured by the Digital Markets Act and EU antitrust enforcement double & triple jeopardy. That framework doesn't care about Washington's corporate proxy wars.
Consider the double jeopardy foreign partnerships face. Under the DMA, the European Commission designates gatekeepers and opens up search data vaults, forcing companies to share proprietary training data with rivals. For a firm like Google, this means giving up the keys to the castle. Or look at French challenger Mistral AI's sovereign ambitions, which thrive because European businesses want models that don't feed into American clouds or answer to US executive orders.
So as Anthropic CEO Dario Amodei warns about the dangers of scaling, he is accused of erecting regulatory moats. David Sacks, the White House AI czar, slammed Anthropic’s policy push as a "sophisticated regulatory capture strategy based on fear-mongering." The Trump administration has pushed back, signing an executive order aimed at stripping power from state-level safeguards. This leaves US startups vulnerable. They are caught between a hostile federal executive at home and the double jeopardy of the DMA abroad.
Midterm Battlegrounds and National Security Stakes
The cash isn't just sitting in PAC bank accounts. It is already hitting candidate battlegrounds. Public First Action launched six-figure ad buys targeting high-profile candidates like Senators Marsha Blackburn of Tennessee and Pete Ricketts of Nebraska. Blackburn has made child online safety a cornerstone of her platform, while Ricketts has championed strict export controls on advanced semiconductors to China.
Anthropic claims this policy push is a direct response to tangible national security threats. They point to the Claude Mythos Preview model, which discovered thousands of high-security vulnerabilities across major browsers and operating systems earlier this year. Anthropic released this intelligence to cyber defenders through Project Glasswing, using it to show that frontier models can be active security liabilities in the wrong hands.
But there's a sharper edge to this story. The federal government previously shut down two of Anthropic’s models due to national security concerns, sparking a lawsuit from the company. The military labeling Anthropic a risk shows that even pro-regulation companies aren't safe from the federal security apparatus. When the state can shut down your models overnight, spending $40 million to educate the public and shape the legislative branch isn't just corporate citizenship. It is an act of survival.