Quarterly lobbying disclosures hit the database Monday. Anthropic's number: $1.6 million. That's 4.4 times what the company spent in the same quarter a year ago. It's also 60% more than OpenAI — the company everyone assumes is the political operator in this race — spent across the same three months.

Different story on the same page: OpenAI logged $1 million. Up from $560,000 in Q1 2025. A respectable bump, but not the curve Anthropic is running.

Both figures are company records.

What they were buying

The disclosure forms list the topics a lobbyist pressed on Capitol Hill and federal agencies. Anthropic's Q1 slate reads like a Department of Defense procurement briefing with an AI sticker on top: "AI procurement," "Defense Department procurement," "supply chain risk," "acceptable use policy," national security, export controls, AI legislation, energy infrastructure, permitting.

"Acceptable use policy" is the detail that actually matters. That phrase exists because Anthropic spent Q1 in a fight with the Pentagon over what classified uses of Claude the company is willing to allow. The company drew red lines. The Pentagon pushed back. That fight — reportedly now in court — is why the lobbying line item grew 340% year-over-year.

Meta, to set the scale, topped all of Big Tech at $7.1 million for the quarter. Anthropic and OpenAI are nowhere near that number yet. But the rate of change is the part venture capitalists read first.

The safety company has the biggest checkbook

Anthropic has spent four years selling itself as the AI lab that chose principles over speed. That positioning is the entire pitch deck. It's how they raised $30 billion, how they recruited defectors from OpenAI, how they convinced enterprise buyers that paying a premium for Claude was a governance decision and not a product decision.

The lobbying filings say something uncomfortable about that pitch. When the safety company is the biggest spender in Washington on AI — bigger than the company it was founded to be safer than — safety stops looking like a philosophy and starts looking like a moat. A regulatory moat, paid for in quarterly installments.

My Opinion

I'll be blunt. Anthropic didn't outspend OpenAI on lobbying because the "acceptable use policy" needs more nuance. They outspent them because "safety" is the cheapest competitive advantage the AI industry has ever invented, and the way you protect that advantage is to write it into federal procurement rules before anyone else can.

The Pentagon fight is genuinely interesting — I don't hate that a US AI lab is drawing classified-use red lines in 2026, and someone should. But the rest of Anthropic's Q1 lobbying spend wasn't about red lines. It was about AI procurement, export controls, permitting, energy — the full list of things that determine who gets to build the next model cycle at scale. That's not safety work. That's incumbency work, dressed up in a lab coat.

The number I'm watching now is Q2. If Anthropic's lobbying spend keeps 4x'ing year-over-year, by 2027 they'll be a rounding error away from Meta's entire Big Tech line. At that point we should probably stop calling them a lab.

Author: Yahor Kamarou (Mark) / www.humai.blog / 22 Apr 2026