A Democratic congresswoman from Massachusetts sat down with a Republican colleague and wrote an AI bill together. That used to be called legislating. In 2025, it gets you a six-figure attack campaign from your own ideological neighborhood.

Americans for Responsible Innovation, an AI safety advocacy group, is rolling out ads this week targeting Rep. Lori Trahan, arguing that her bipartisan work with Rep. Jay Obernolte (R-California) amounts to siding with what the group calls AI oligarchs. The campaign is explicitly aimed at pressuring Trahan to walk back her support for legislation that would temporarily preempt state AI regulations while a federal framework is worked out. The group is not running ads against Obernolte.

The Trahan-Obernolte draft bill proposes a three-year moratorium on state-level AI regulation — meaning California's AI laws, New York's emerging rules, and dozens of other state-level efforts would be frozen while Congress theoretically gets its act together. That three-year window is the central flash point. Supporters call it necessary breathing room to avoid a 50-state patchwork of contradictory rules that would strangle AI development. Critics call it a gift to the industry: three years of regulatory vacuum, delivered gift-wrapped by Congress, with no guarantee a federal replacement ever materializes.

The criticism from Americans for Responsible Innovation lands in a specific political context that the attack ads benefit from not having to explain fully. The AI safety community — a loose coalition of researchers, ethicists, and advocacy organizations that believes advanced AI poses serious societal risks — has a problem right now that goes beyond any single bill: they don't trust the Trump administration to write AI regulation. The concern is structural. Handing preemptive authority over AI governance to a federal government currently run by people who have made deregulation a core identity means that the three-year pause could function, in practice, as permanent deregulation.

That fear is legitimate and worth taking seriously. It is also convenient. The same AI safety lobby that frames itself as standing against powerful tech interests is, in this instance, functionally defending the status quo: a fragmented, state-by-state approach that the largest AI companies — with the legal firepower to navigate 50 different jurisdictions — are better positioned to survive than any startup or civil society group trying to hold them accountable. The politics of AI regulation are scrambled enough that both sides can credibly claim the other is carrying water for Big Tech.

What the draft bill actually says matters more than what either side says about it. The Trahan-Obernolte framework, as released, would prohibit states from enforcing AI-specific regulations on AI model development during the preemption window. It does not eliminate federal agency authority, though the scope of that authority under the current administration is itself contested. The bill does not appear to include enforceable federal AI safety standards that would replace what the states are being asked to stand down from — which is the core substantive critique, and a fair one.

Trahan has defended the approach as a necessary first step toward a coherent national framework, arguing that legislative inaction is itself a choice with consequences. Obernolte, a Republican who has engaged seriously with AI policy for longer than most of his House colleagues, has made a similar case. Both have positioned the bill as an opening bid, not a final answer — standard legislative framing that is either honest or optimistic depending on how much faith one has in Congress's follow-through.

Meanwhile, Anthropic — the AI safety-focused company that nonetheless operates one of the most powerful AI systems in the world — is reported to be filing for a blockbuster initial public offering. That detail sits in the background of this entire fight and deserves not to be background. The company most publicly associated with the idea that AI development is dangerous enough to require serious constraints is also preparing to go public at a valuation that would make its founders and investors extraordinarily wealthy. The AI safety movement and the AI industry are not, in 2025, cleanly separable entities.

The attack on Trahan may succeed politically. Six-figure ad campaigns in a Democratic primary context can move numbers. But the more significant story is what the campaign reveals about how AI governance is actually being fought: not in committee hearings or regulatory dockets, but in the same campaign-ad machinery that governs every other Washington fight, funded by interests that have their own complicated relationship with the technology they claim to want to control.