Steelman Analysis

Our Thesis (Restated)

The AI industry has invented "bilateral capture" -- the systematic funding of candidates in both parties' primaries via dual-PAC architectures, running ads about everything except AI, to ensure that regardless of which party wins any given seat, the winner was pre-approved by the industry, effectively privatizing candidate selection and making elections irrelevant to the question of AI regulation.

Primary Counterargument

The article conflates sophisticated political engagement with antidemocratic manipulation, and the "bilateral capture" framework overstates both the novelty and the efficacy of what's happening.

The strongest version of this objection comes from political scientists who study campaign finance empirically rather than narratively. The core claim of the thesis -- that bilateral PAC spending effectively "pre-decides" elections by curating the candidate menu -- rests on an assumption that money reliably determines primary outcomes. The political science evidence on this is genuinely mixed. Research from the University of Arizona finds that super PACs have "muted equilibrium effects" on elections because political competition constrains their impact: when one side spends, the other side responds, and the net effect on outcomes is smaller than either side's gross spending would suggest. David Primo (Rochester) and Jeffrey Milyo (Missouri), who have aggregated decades of survey and electoral data, argue that Americans systematically overestimate how much campaign spending determines results, and that changes in campaign finance rules have "little to no effect" on outcomes. Bloomberg's $1 billion presidential flameout is the most dramatic data point, but the pattern holds more broadly.

The thesis would respond that primaries are different from generals -- lower turnout, less voter information, greater marginal impact per dollar. That's true, and it's the thesis's strongest empirical ground. But even in primaries, the relationship between spending and winning is not deterministic. AIPAC spent $14.5 million to defeat Summer Lee in Pennsylvania's 2024 primary and failed. The crypto-backed candidate who lost to Katie Porter in 2024 had massive PAC support and still lost. If bilateral capture were as mechanically effective as the framework implies, these failures wouldn't happen. The thesis needs to grapple with the fact that its central mechanism -- spending controls who gets nominated -- is a tendency, not a law, and the framework's rhetorical power ("democracy's back end has been captured") oversells the certainty of the outcome.

More fundamentally, the "bilateral capture" framework assumes that industry-funded candidates arrive in Congress as captured agents who will do the industry's bidding. But the causal chain from "received PAC support" to "votes the way the PAC wants" is weaker than the thesis acknowledges. Members of Congress have multiple donors, multiple constituencies, and their own ideological commitments. The thesis treats PAC spending as a purchase when the empirical literature suggests it's closer to a bet -- industries support candidates who already lean their way, then claim credit when those candidates do what they were going to do anyway. The direction of causation matters enormously, and the thesis doesn't adequately address it.

Who Makes This Argument

This is the argument of libertarian-leaning campaign finance scholars (Bradley Smith, founder of the Institute for Free Speech; David Primo and Jeffrey Milyo), conservative and centrist legal scholars who defend Citizens United on First Amendment grounds, and political scientists in the "money doesn't buy elections" school (Steve Levitt's early work, though he's more nuanced now). It's also the implicit position of the AI industry itself -- Leading the Future spokesperson Jesse Hunt frames the spending as advocacy for American competitiveness, not as an attempt to capture Congress.

Why It Has Merit

The political science evidence genuinely does not support a simple "money buys outcomes" story. The relationship between campaign spending and electoral results is real but noisy, moderated by candidate quality, partisan lean, incumbency, and competitive dynamics. The thesis would be stronger if it acknowledged this uncertainty rather than presenting bilateral capture as a near-mechanical process. The framework's power comes from its elegance -- "fund both doors, win either way" -- but elegance is not evidence of efficacy. Some PAC-backed candidates will lose. Some who win will vote against the industry on key bills. The thesis needs to be honest about the gap between the architectural intent (which is real and well-documented in FEC filings) and the realized policy outcome (which is probabilistic, not guaranteed).

Additionally, the empirical finding that super PAC spending can benefit challengers over incumbents -- with freshman congressional members averaging 79 per cycle after Citizens United versus 55 before -- complicates the narrative that outside spending is inherently antidemocratic. If PAC money helps newcomers unseat entrenched incumbents, the democratic valence isn't as one-directional as the thesis implies.

Where It Falls Short

This counterargument ultimately confuses "money doesn't guarantee outcomes" with "money doesn't matter." The fact that PAC spending is probabilistic rather than deterministic doesn't mean the structural pattern is democratically harmless. You don't need a 100% success rate for bilateral capture to reshape the political landscape -- you need a high enough success rate to change the expected value calculation for candidates considering pro-regulation positions. The Fairshake precedent is the strongest rebuttal: the crypto PAC's 2024 bilateral spending defeated Sherrod Brown, helped pass the GENIUS Act stablecoin legislation, and installed industry-friendly regulators. That's not a theoretical pattern; it's a documented causal chain from spending to personnel to policy. The political science literature on aggregate PAC effectiveness doesn't adequately account for the specific dynamics of low-turnout primaries where a few million dollars can constitute a majority of all spending in the race.

Moreover, the "money follows ideology, not the other way around" argument has a selection-effect problem of its own. If the industry is systematically identifying and funding candidates who already lean anti-regulation in both parties' primaries -- which is exactly what the dual-PAC architecture is designed to do -- then the fact that those candidates "already believed" what the industry wants doesn't make the spending innocuous. It means the industry is tilting the primary playing field toward pre-existing anti-regulation candidates and against pro-regulation ones. The capture isn't of individual minds; it's of the candidate pool.

Secondary Counterarguments

The Legitimate Policy Argument: State Patchwork Is a Real Problem

The AI industry's stated rationale for political spending -- opposing a patchwork of contradictory state regulations in favor of a coherent federal framework -- is not a pretextual smokescreen. It's a genuine policy concern with real evidence behind it. In 2025, state legislators introduced over 1,200 AI-related bills, with at least 145 becoming law. These laws impose contradictory requirements: a single hiring tool must simultaneously satisfy California's recordkeeping standards, Colorado's impact assessments, and New York City's bias audit mandates. The startup PerceptIn budgeted $10,000 for compliance and spent $344,000 -- more than double its R&D costs -- before shutting down. Harvard Kennedy School researchers identified a "compliance trap" where regulatory costs consume resources faster than startups can generate revenue, transforming operating margins from positive 13% to negative 7%. The U.S. Chamber of Commerce estimates that applying Colorado-style requirements nationwide could cost 92,000 jobs.

This matters for the article because the thesis treats the industry's stated preference for federal over state regulation as a euphemism for "no regulation." But preferring federal uniformity over state fragmentation is a substantively defensible position held by many good-faith policy actors, and it's the same position the U.S. has taken on financial regulation, pharmaceutical approval, and telecommunications. The article needs to distinguish between the industry's policy preference (which has real merit) and its political method (which is the actual problem). Conflating the two makes the argument easier to dismiss.

Assessment: This is the counterargument most likely to resonate with the article's target audience of center-left, policy-literate readers. Many of them instinctively favor federal coherence over state fragmentation in other policy domains. The article should acknowledge the patchwork problem as real while making clear that the appropriate response is federal legislation arrived at through democratic deliberation -- not industry pre-selection of the legislators who will write that legislation.

The First Amendment Defense: Legal Speech in a Free Society

Super PAC spending is constitutionally protected political speech under Citizens United v. FEC (2010). This isn't a technicality; it reflects a substantive First Amendment principle that the government cannot restrict political speech based on the identity of the speaker or the amount of money spent. As Justice Kennedy wrote for the majority, restricting independent expenditures "equates to limiting their speech." The Institute for Free Speech argues that Citizens United has demonstrably expanded political competition, noting that freshmen in Congress increased significantly after the decision, and that outside spending disproportionately benefits challengers over incumbents.

The deeper version of this argument isn't just "it's legal, deal with it." It's that any framework powerful enough to prevent bilateral capture would necessarily restrict political speech in ways that create their own democratic dangers. Who decides which spending is "capture" and which is "advocacy"? Environmental groups spend in both parties' primaries too. Labor unions support candidates based on labor policy. The bilateral spending pattern the thesis identifies is structurally identical to what many progressive advocacy groups do. If you build a legal or rhetorical framework that delegitimizes the AI industry's bipartisan primary spending, you've also delegitimized the AFL-CIO's, the Sierra Club's, and Planned Parenthood's.

Assessment: The thesis already flags this as the strongest counterargument, which is correct. The article should engage it by drawing the line at scale and asymmetry rather than at the act of spending itself. The problem isn't that industries participate in primaries -- it's that $190M+ in anti-regulation money versus $50M in pro-regulation money, deployed through deliberately opaque dual-PAC structures running ads about unrelated issues, distorts the primary process at a scale that no advocacy group matches. The quantitative difference becomes a qualitative one. But the article must be precise here, because sloppy versions of this argument do slide into restricting legitimate advocacy.

The Industry Isn't Monolithic: Anthropic and the Internal Debate

The thesis's framing of a unified industry engineering a captured Congress is complicated by the fact that Anthropic -- one of the leading AI companies -- has put $20 million into Public First, a pro-regulation super PAC, and is actively working to elect candidates who support stricter AI oversight. Public First itself is run by bipartisan former members of Congress (Republican Chris Stewart and Democrat Brad Carson) and is spending in both parties' primaries to support pro-regulation candidates. If the AI industry were executing a coordinated capture strategy, one of its most prominent members wouldn't be spending tens of millions fighting the other side.

This undermines the thesis's rhetorical framing of "the industry" as a monolithic actor. The real picture is an intra-industry policy war, with anti-regulation forces (OpenAI ecosystem, a16z, Palantir, Meta) outspending pro-regulation forces (Anthropic) by roughly 4:1 to 6:1. That spending asymmetry is still concerning, but it's a different story than "the industry is buying Congress." It's closer to "the anti-regulation faction of the industry is outspending the pro-regulation faction" -- a less narratively clean but more accurate description.

Assessment: The thesis already notes this complication, which is good. But it should resist the temptation to dismiss it as merely a 6:1 spending ratio. The existence of Anthropic's counter-spending matters because it demonstrates that the AI industry could engage in political spending that supports regulation. The relevant question isn't whether spending exists but whether the market for political influence is so lopsided that pro-regulation voices can't compete. The article should frame Anthropic as evidence that the bilateral capture strategy is a choice made by specific actors, not an inevitable feature of industry political participation.

The "Nothing New Under the Sun" Objection

Industries have always spent money on politicians who favor their interests. Railroads did it in the Gilded Age. Oil companies did it throughout the 20th century. Pharmaceutical companies, tobacco, Wall Street -- every major industry has cultivated political allies in both parties. The thesis's claim that bilateral capture is "categorically different" from traditional lobbying rests on the specifics of dual-PAC architecture and primary targeting, but these are differences of degree and sophistication, not of kind. Calling it a new concept risks overstating the novelty and understating the continuity of American plutocratic politics.

Assessment: This objection has real force but ultimately strengthens rather than undermines the thesis if handled correctly. The article should acknowledge the continuity -- bilateral capture is the latest evolution of a long American tradition of industry political influence -- while insisting that the specific innovations matter. The deliberate construction of parallel PACs for each party's primaries, the systematic misdirection of ad content away from the industry's policy agenda, and the replication of a proven playbook (Fairshake) across industries represent a qualitative escalation in the precision and effectiveness of capture. The Gilded Age railroads didn't have FEC-optimized dual-PAC structures and A/B-tested attack ads about immigration designed to obscure the real policy transaction. The mechanism has been industrialized.

Our Weak Points

1. The elections haven't happened yet. The thesis is making a structural argument about a process that is still in progress. We can document the spending architecture and the intent, but we cannot yet demonstrate the outcome. If AI-backed candidates lose key primaries, or if winners vote for regulation despite industry support, the bilateral capture framework will look like an overreach. The Fairshake precedent is the best evidence the strategy works, but it's a single case from a different industry. The article should be explicit about this epistemic limitation.

2. The causal chain from spending to policy is asserted more than demonstrated. The thesis's strongest evidence of the "harvest" stage is the Hochul RAISE Act gutting and the Trump executive order. But both of these happened through executive action, not through Congress. The article claims bilateral capture is about building a Congress that can't regulate, but its best examples of policy capture don't involve Congress at all. This is a structural weakness in the argument that opponents will exploit.

3. The "misdirection" framing assumes bad faith that may not be accurate. The thesis treats the fact that AI PAC ads focus on immigration, healthcare, and Trump rather than AI as evidence of deliberate concealment. But as both Leading the Future and Public First acknowledged, AI regulation is simply not a salient voter issue yet. Running ads about what voters care about is standard political communication, not misdirection. Every issue PAC does this -- environmental groups run ads about jobs, not carbon pricing; gun control groups run ads about school safety, not Second Amendment jurisprudence. The article needs to explain why this is structurally different rather than just rhetorically suspicious.

4. The 69-80% public support for AI regulation is softer than it looks. Polling consistently shows large majorities favoring AI regulation in the abstract, but "regulation" is a vague concept that encompasses everything from light disclosure requirements to development moratoriums. When pollsters ask about specific regulatory proposals, support fractures along partisan and demographic lines. The article uses these topline numbers to argue the industry is fighting against the public will, but the public's will on AI regulation is far less coherent than the polling suggests.

5. The framework may not travel as well as claimed. The thesis wants "bilateral capture" to become a general-purpose concept applicable to crypto (Fairshake), pro-Israel politics (AIPAC), and future industries. But the structural parallels may be more superficial than substantive. AIPAC's influence operates through a dense network of grassroots relationships, bundled individual donations, and decades of institutional infrastructure -- not just PAC spending. Crypto's political spending is more structurally similar but targeted different policy outcomes (permissive regulatory frameworks vs. blocking regulation entirely). Lumping these together under one concept risks sacrificing analytical precision for rhetorical punch.

Recommended Handling

Must address head-on (dedicate significant space):

• The political science evidence on spending effectiveness. The article already plans a steelman section, and this should be its centerpiece. Acknowledge that money doesn't mechanically determine outcomes. Then make the case that bilateral capture works at the level of expected value -- it shifts the probability distribution of who gets nominated, and in low-turnout primaries, that shift is large enough to matter. The Fairshake precedent is the key evidence.
• The legitimate patchwork problem. Concede that state regulatory fragmentation is a real compliance burden, especially for startups. Then draw the sharp distinction: the appropriate response to fragmentation is democratic deliberation over a federal framework, not industry pre-selection of the legislators who will write it. The method is the problem, not the policy preference.

Acknowledge and engage briefly:

• The First Amendment defense. State clearly that the spending is legal and constitutionally protected. Then make the "legal is not democratic" argument efficiently. This is well-trodden ground; don't belabor it.
• The Anthropic counter-spending. Note it, frame it as evidence that capture is a choice not an inevitability, but don't let it dilute the structural argument. The 4:1 to 6:1 spending ratio tells the story.

Proactively raise before critics do:

• The "nothing new" objection. The article should own the historical continuity early -- "industries have always bought influence" -- and then use it as a springboard to explain what's specifically new about the dual-PAC primary architecture. This inoculates against the accusation of conceptual novelty for its own sake.
• The epistemic limitation that the elections haven't happened. Frame the article as naming the strategy in real time, before the outcome is known, which is precisely when it can still be made politically costly. Waiting for results means waiting until the mechanism has already worked.

Watch for but don't overinvest in:

• The "misdirection" framing. This is the thesis's most rhetorically appealing but analytically vulnerable claim. The article should present the ad content pattern as evidence of what the money is and isn't buying, but avoid language that implies deliberate voter deception unless the evidence supports that specific intent. "The ads are about what voters care about, not what the money wants" is more defensible than "the ads are designed to hide the real agenda."