One in four federal lobbyists now works on artificial intelligence. The public debate focuses on deepfakes. The structural reshaping of governance happens in lobbying disclosures nobody reads.
Public Citizen published its annual lobbying census in February 2026. The headline number should have reordered every conversation about AI regulation: 3,570 federal lobbyists worked on artificial intelligence issues in 2025. That is one in four of all registered lobbyists in Washington. The count grew 168 percent since 2022. Measured by lobbyist-client relationships, the growth was 265 percent, from 1,672 pairings to 6,110.
The number did not reorder the conversation. The conversation was elsewhere.
The Decoy
A study presented at the ACM Web Conference in 2026 tracked 187,778 social media posts during the Canadian federal election in late April 2025. Deepfakes appeared in 5.86 percent of election-related images. The harmful subset accounted for 0.12 percent of all views on X. The Harvard Ash Center reviewed elections across more than eighty countries in 2024 and found no documented case where a deepfake changed an electoral outcome. The Biden robocall in New Hampshire was investigated and prosecuted. Fake celebrity endorsements circulated and dissipated. AI-generated hurricane images spread and were debunked.
Deepfakes are real. Their demonstrated impact on elections is, so far, zero. The regulatory energy devoted to them is enormous. The EU AI Act contains extensive deepfake disclosure requirements. Twenty-three US states passed deepfake-related legislation in 2024 alone. Congressional hearings have featured deepfake demonstrations and testimony from victims. None of this attention is wasted in principle. But attention is a finite resource, and every hour spent on deepfake hearings is an hour not spent examining what 3,570 lobbyists are actually doing.
The Structural Reality
Meta spent $26.3 million on federal lobbying in 2025, more than any other individual company in any sector. Nvidia spent $4.9 million, seven times its 2024 outlay. Anthropic outspent OpenAI for the first time in the first quarter of 2026, spending $1.6 million to OpenAI's $1 million. The total AI industry federal lobbying spend reached approximately $105 million. Ninety-one of the top hundred AI lobbying entities are corporations or corporate trade associations.
The spending is concentrated and directional. Data center lobbying alone grew nearly five hundred percent. The US Chamber of Commerce deployed 91 lobbyists on AI issues, Microsoft 63, Meta 55, Intuit 51, Amazon 48. These numbers describe an industry that has learned from the pharmaceutical and financial services playbooks: establish the regulatory framework before the public understands what is being regulated.
The lobbying produced a concrete result. On December 11, 2025, the White House signed an executive order directing the Department of Justice to create an AI Litigation Task Force empowered to sue states over AI laws. The same order conditioned BEAD broadband funding on state alignment with federal AI policy and directed the drafting of federal preemption legislation. Congress has passed no comprehensive AI regulation. The executive branch, guided by industry lobbying, is building the regulatory architecture by default.
The Attention Asymmetry
The information structure of deepfakes makes them irresistible to media and legislators. They are visual, visceral, and easy to demonstrate in a hearing room. A senator can hold up a deepfake image and ask whether democracy can survive this threat. The question writes its own headline.
Lobbying disclosure filings are the opposite. They are dense, delayed, and require sustained attention to interpret. The Public Citizen report that documented the 3,570-lobbyist figure received a fraction of the coverage that any single viral deepfake generates. The ratio of public attention to demonstrated impact is inverted: deepfakes attract high attention with low documented harm, while lobbying attracts low attention with high structural consequence.
This is not a conspiracy. It is an information market operating as designed. Compressible narratives win attention. A deepfake can be explained in a screenshot. A lobbying campaign that produced an executive order preempting state AI laws requires reading a 2,400-word executive order, cross-referencing lobbying disclosures, and tracking the legislative calendar. The complexity is the cover.
Who Wins
The companies spending $105 million annually on AI lobbying are winning. The executive order they helped produce will, if its preemption framework holds, prevent states from passing restrictive AI legislation, consolidating regulatory authority at the federal level where industry influence is most concentrated.
The losers are the state legislatures that passed more than 1,500 AI-related bills in 2024, many of which addressed concerns their constituents raised. Federal preemption does not replace state regulation with federal regulation. It replaces regulation with the absence of regulation, because the federal legislation that preemption assumes has not been written.
The deepfake threat is real but contained. The lobbying saturation is real and compounding. The attention of the people who write the rules is finite. That is the misdirection.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)