Summary

  • AI industry super PACs have deployed roughly $44 million in 2026 midterm spending, with nearly half concentrated on the Democratic primary in New York’s 12th Congressional District, functioning as what Public First founder Brad Carson has characterized as a test of whether concentrated industry money can neutralize electoral consequences for pro-regulation candidates.
  • Leading the Future, a pro-AI super PAC network funded by a $75 million war chest from four donors — venture capitalists Marc Andreessen and Ben Horowitz, and OpenAI co-founder Greg Brockman with his wife Anna — has spent $8.2 million targeting state Assemblymember Alex Bores, who authored New York’s Raise Act requiring major AI developers to publish public safety plans.
  • A safety-advocacy coalition including You Can Push Back, Public First, and the newly launched Guardrails Alliance has spent $11 million in the same race, though the coalition’s funding flows partly through a dark-money intermediary not required to disclose all donors, creating a transparency asymmetry the article does not resolve.
  • The contest’s dominant framing as an “AI civil war” naturalizes a zero-sum contest between technological acceleration and precautionary regulation while obscuring shared stakeholder interests in workforce transition and the structural absence of federal AI legislation as a causal driver of the spending dynamic.

The Democratic primary in New York’s 12th Congressional District has become the most expensive single contest in what has been described as the AI industry’s first concentrated campaign spending cycle. AI-focused super PACs have raised roughly $100 million this cycle, of which approximately $44 million has been spent, with nearly half flowing into one House race between two state assemblymembers who both campaign on AI guardrails. The race’s significance rests not only on who wins but on what the spending pattern reveals about the political economy of artificial intelligence regulation at a moment when public opinion runs sharply against the industry’s preferred pace of deployment.

The Money

The primary’s spending dynamics center on two PAC networks operating in opposition. Leading the Future, a bipartisan network of super PACs created to back pro-AI candidates, has spent $8.2 million on the race, according to The Guardian’s reporting, which cites FEC data. The group is funded by a $75 million war chest drawn from four donors: venture capitalists Marc Andreessen and Ben Horowitz, and OpenAI co-founder Greg Brockman with his wife Anna. The message the PAC network was sending, Carson told The Guardian, was: “Regulate AI, and we will find you, wherever you are.” Leading the Future did not respond to The Guardian’s request for comment.

On the other side, a coalition of super PACs advocating for stronger AI safeguards has spent $11 million in the same contest. You Can Push Back is funded by crypto billionaire Chris Larsen. Public First, whose Democrat-focused subsidiary is Jobs and Democracy, was founded by Carson, a former Democratic congressman from Oklahoma. Anthropic has publicly announced a $20 million contribution to Public First, and Carson said the organization has raised another $45 million from “various industries, including people working at major AI labs.” Public First’s funding is partially opaque: it operates through a dark-money group not required to disclose all donors. On Thursday, June 22, Guardrails Alliance launched as a new AI-focused super PAC explicitly built to counter Leading the Future, with backing from several labor unions and Chris Hyams, the former Indeed CEO who stepped down last year over AI concerns. A spokesperson said the organization will not accept corporate money.

The $44 million in total AI-focused super PAC spending this cycle, with nearly half concentrated in a single district, signals that both sides have designated the race as the defining electoral test of the 2026 cycle on AI policy.

The Candidate

State Assemblymember Alex Bores, a former tech worker, authored the Raise Act — described as the second state law in the U.S. requiring major AI developers to publish public safety plans. Bores entered the race as an underdog but has emerged as the central figure in the spending contest. His framing of the primary as a referendum on the industry’s influence is encapsulated in his campaign video statement: “This is the first congressional race in the country where the dividing line is: can we regulate AI at all?”

Polls now show Bores in a tight race with fellow Assemblymember Micah Lasher, who has also campaigned in favor of AI guardrails and curbing Big Tech power. Lasher occupies a parallel but distinct position: he shares the guardrails platform without the symbolic weight of Raise Act authorship, and his interest in distinguishing his candidacy while not being branded as the PAC-preferred alternative represents a procedural dynamic the article does not explore. That both candidates share a broadly similar policy orientation on AI while one absorbs the overwhelming majority of targeted PAC spending is itself analytically significant — the money is not buying a policy difference so much as sending a signal about the costs of authorship.

Beneath the Positions

The spending maps onto interests that extend well beyond stated policy positions.

Leading the Future’s four donors combine economic interests with identity interests. Andreessen and Horowitz hold venture portfolios heavily concentrated in AI investments; the $75 million check size relative to other 2026 spending suggests the regulatory outcome matters substantially to concentrated capital, not merely to diffuse industry sentiment. The Brockmans’ inferred interest is more ambiguous: they may be driven by conviction that rapid AI advancement is a public good and regulation is a decelerant, but without direct confirmation this remains a hypothesis.

The strategic effect of Leading the Future’s spending, per Carson’s characterization, is to communicate to individual legislators that pursuing safety regulation invites costly electoral opposition. The safety coalition’s $11 million spend, in parallel, seeks to demonstrate that pro-safety platforms remain electorally viable, thereby encouraging other legislators to adopt similar positions without fearing political extinction.

Anthropic’s confirmed $20 million contribution to Public First — the company’s own messaging describes this as support for “targeted regulation focused on the nearest-term high risks: AI-enabled biological weapons and cyberattacks” — raises a separate analytical question. Whether a commitment to safety regulation also yields a competitive advantage for an established firm whose deployment model is slower than the headline pace — a dynamic sometimes characterized as “pulling up the ladder” where regulation disproportionately burdens new entrants — is a hypothesis no external source attributes to competitive-strategy motives. The question remains unverified.

Carson’s statement that $45 million came from “various industries, including people working at major AI labs” suggests internal industry dynamics are a significant driver of the safety coalition, not merely public-interest advocacy. Whether those funds are predominantly from AI-adjacent workers motivated by labor interests, competing AI firms motivated by commercial strategy, or broader-affiliated technology philanthropists shapes whether the safeguard side represents a genuine public-interest counterweight or a different-faction political operation. The article does not resolve this.

The safety coalition’s financing is itself partly opaque. Public First’s primary funder is a dark-money group not required to disclose donors, and the $45 million from “various industries” could include interests whose alignment with safety regulation is contingent on the specific regulatory outcome. Larsen’s crypto-industry interest may include ensuring AI development does not destabilize blockchain-dependent assets. Labor unions’ backing of Guardrails Alliance reflects workforce anxieties about displacement, concern about workplace AI surveillance, and institutional self-preservation in a chosen political coalition. Hyams’ departure from Indeed “over AI concerns” and subsequent PAC formation suggests an identity-driven component — moral positioning about AI’s pace that functions as a signaling act for other executives who believe faster development threatens their reputational standing.

Bores’ interests are dual: substantive, in that the Raise Act represents career-defining legislative work, and identity-driven, in that national-figure positioning depends on the race being understood as “the first congressional race in the country where the dividing line is: can we regulate AI at all.” His framing reflects a strategic calculation that constituent interest in AI regulation will translate into electoral support, a bet whose validity depends on voter-opinion data the article grounds in county-level employment figures and a national poll rather than district-specific surveying.

The spending asymmetry complicates the article’s underdog characterization. Leading the Future’s $8.2 million in the primary is offset by the safeguard coalition’s $11 million in the same race, and the full $75 million war chest is matched by a partially disclosed total for Public First with a floor of roughly $65 million. The resource balance is not as lopsided as the narrative framing implies.

A confirmed, shared interest across many actors — workforce transition stability — is obscured by the zero-sum frame. The Brookings Institution has named New York City the nation’s most “AI-exposed” county, where one-fifth of the workforce holds jobs AI could plausibly replace — predominantly white-collar roles such as software developers, marketers, and financial analysts. The unions’ involvement and the “Jobs and Democracy” subsidiary name both point to a widely held need for policies mitigating AI-induced labor disruption. An integrative compact pairing safety-reporting requirements with public-private reskilling investment and tax-advantaged retraining models would align labor, tech firms seeking social license, and safety-conscious AI developers. That possibility remains unthinkable inside the “civil war” frame.

The genuinely oppositional interest is temporal — how quickly regulation should be imposed — which is inherently distributive but need not be absolute if institutional design incorporates phased thresholds. No stakeholder in this race has endorsed such an approach.

How the Contest Is Being Framed

The dominant frame around the race has been set by Carson’s characterization of the spending dynamic as “the AI civil war.” The metaphor naturalizes a zero-sum contest between two wings of the AI industry — those favoring rapid deployment and those favoring precautionary regulation — while positioning the primary as symbolic terrain for a larger struggle. By framing the contest as intra-industry, the metaphor obscures wider public and regulatory actors.

Per Robert Entman’s framing formulation, the article’s construction operates across four dimensions. The problem is defined as AI deployment as an emergent threat to livelihoods and democratic accountability, with the primary as symbolic terrain. The causal interpretation foregrounds wealthy tech donors spending to defeat regulation-supporting candidates, carrying what Carson characterized as an explicit threat — “Regulate AI, and we will find you, wherever you are.” The moral evaluation repeatedly evokes elite overreach, through Henry Ajder’s observation about “the dynamics of Wall Street and the opaque sense of elites making decisions about us that don’t benefit us” and through the YouGov data showing two-thirds of voters believe AI is advancing too quickly and only one in five think its economic impact will be positive overall — views held evenly across party lines. The treatment recommendation frames the primary as a “referendum” on whether AI can be regulated “at all.”

War-metaphor vocabulary throughout the article — “battleground,” “proxy fight,” “counter-assault,” “blitz,” “blanketing” — renders the spending dynamic as armed conflict and assigns Bores a protagonist role. The central policy metaphor of “guardrails” lexicalizes regulation as physical safety infrastructure, activating protection schemas rather than innovation schemas. Carson’s direct quote can be interpreted as casting pro-acceleration forces in a predatory light. The Brookings Institution’s phrase characterizing AI-exposed counties as “hotbeds for some of the AI era’s most agitated voters” can similarly be read as evoking a disease-like metaphor for public anxiety. Both are interpretive readings of the source’s language, not descriptive facts about the source’s intent.

The article’s sourcing architecture privileges quotations framing spending as aggressive. Carson’s threatening characterization constitutes one of only two direct quotes attributed to a named PAC principal in the article. Ajder’s closing observation — that “AI companies are increasingly being seen in a similar light, whether you’re on the right or the left” — activates a class-resentment frame. Ajder also noted that even the most cautious AI executives face a “constant pressure to release new models quickly,” creating an inherent tension with calls for slower development.

Nominalizations throughout the article — “the dynamics of Wall Street,” “the opaque sense of elites making decisions” — remove human agency, presenting systemic forces rather than individual choices. The article names donors on both sides. The asymmetry lies in funding-channel transparency: the safeguard side’s funding flows through a dark-money intermediary not required to disclose all donors.

A counterframe of institutional mechanism — one that foregrounds the absence of federal legislation as a causal story, shifts moral evaluation toward democratic-process design, and pivots treatment recommendations toward evidence-based policy — remains suppressed by the zero-sum frame’s dominance. The NY-12 primary is both a product of that framing struggle and a potential pivot point for what becomes thinkable next.

What the Primary Could Set in Motion

Several structural conditions shape whatever outcome Tuesday produces. AI capability will continue advancing. The New York metro area’s labor market will face accelerating white-collar exposure. Public apprehension will remain elevated regardless of the primary’s result. These are predetermined elements. The Brookings vulnerability data and the YouGov skepticism finding — two-thirds of voters believe AI is advancing too quickly, only one in five expect positive economic impact, views distributed evenly across party lines — represent structural political conditions that money alone may not be able to override. The crypto playbook of 2024, when more than $200 million in PAC money helped crypto-aligned candidates win the overwhelming majority of targeted races, succeeded in a different voter-opinion environment: cryptocurrency was not perceived as displacing jobs at scale.

If the safety candidate wins and federal legislation subsequently passes, tech donors are likely to shift from blanket opposition toward shaping the details of regulation, seeking to avoid a state-by-state patchwork that would be navigable by compliance-heavy firms but punitive for smaller labs. Leading indicators for this scenario would include increased bipartisan co-sponsorship of comprehensive AI legislation, donor pledges to respect safety frameworks, and a drop-off in aggressive anti-regulation PAC spending.

If the safety candidate wins but no federal legislation materializes, the result constitutes a symbolic victory demonstrating public-opinion potency while accelerationist donors resume focus on blocking state-level bills, replicating the multi-state playbook across subsequent races. Leading indicators would include safety-aligned PACs shifting resources to other state contests, Congressional AI bills stalling in committee, and Leading the Future maintaining a multi-state strategy targeting local legislators.

If the anti-regulation candidate wins and federal legislation passes, Congress — perhaps alarmed by the spending itself — enacts preemptive guardrails more permissive than safety advocates sought, with industry securing a light-touch framework that preempts stricter state laws. Leading indicators would include disclosure filings showing Leading the Future shifting funds toward lobbying for federal preemption language, corporate executives publicly endorsing a light-touch federal framework, and safety PACs beginning to fracture between those who accept compromise and those who refuse.

If the anti-regulation candidate wins and no federal legislation follows, the signal to state legislators nationwide would be that opposing AI development is politically fatal; safety coalition fundraising would erode; and the default self-regulatory regime would persist with sporadic state backlash failing to coalesce. Leading indicators would include multiple state-level candidates who opposed AI regulation winning primaries with significant PAC support, safety-PAC fundraising declining for two consecutive quarters, and tech firms announcing large-scale hires in exposed counties without accompanying reskilling commitments.

A third possibility operates independently of the candidate outcome. The sheer scale of spending — roughly $44 million concentrated in one House race — may crystallize voter perception of AI-industry political overreach. The latent constituency suggested by the YouGov data becomes visible. The 2026 general election could produce representatives less aligned with the AI industry, not because PACs failed to spend but because spending itself became the issue. The rural data-center dimension — PACs spending across Utah, Texas, Ohio, Georgia, and Kentucky, often against local backlash — suggests this dynamic is already active at district level. Candidates running explicitly against AI-industry PACs post-NY-12, Republican candidates adopting anti-AI-spending rhetoric (the YouGov skepticism data crosses party lines), and state-level AI safety legislation advancing in PAC-active jurisdictions would all serve as leading indicators.

Beyond NY-12, the PAC networks have already spent millions in primaries across multiple states. Public First has supported candidates advocating for AI advancement, including $1.5 million for Texas House candidate Carlos De La Cruz and nearly $1 million for Representative Celeste Maloy of Utah. It has also spent heavily on lawmakers overseeing AI legislation, including $1.6 million behind Representative Valerie Foushee, who co-chairs the House Democratic Commission on AI. The geographic spread signals a multi-state operation extending well beyond the symbolic NY-12 contest.

A wild-card scenario involves industry fragmentation. If a major AI company funding safeguard-advocacy PACs ceases to use intermediaries and states its political position directly — or if opposition to the Raise Act becomes a reputational liability in adjacent business lines — the political alignment could fragment beyond what the current PAC architecture represents. The tension between Anthropic’s stated safety mission and its participation in a dark-money funding structure is one empirical instance of this pressure.

Unresolved Dynamics

The article provides no NY-12-specific public-opinion polling on AI, relying instead on county-level employment data and a national YouGov poll. Whether New York City voters’ material AI exposure translates into the nationally measured skepticism or into a different political response — perhaps one driven more by the local race’s dynamics, including that both candidates campaign on guardrails — is an inference, not a confirmed finding.

Whether the $45 million Carson described as coming from “various industries” is predominantly from AI-adjacent workers motivated by labor interests, from competing AI firms motivated by commercial strategy, or from broader-affiliated technology philanthropists shapes whether the safeguard side represents a genuine public-interest counterweight or a different-faction political operation. Public First’s dark-money intermediary adds a layer the article acknowledges but does not resolve.

Whether the pro-AI PAC network’s strategy is pre-emptive — aimed at freezing the regulatory environment before the 2027 presidential cycle — or reactive, responding to specific state-level legislation like the Raise Act, is not distinguished by the article. Bores’ characterization — “can we regulate AI at all” — implies the latter; the concentrated spending pattern across multiple states argues for the former. These are not mutually exclusive.

Strategic Indicators

For elected officials monitoring AI policy, the position that concentrated PAC spending draws public scrutiny into the existing regulatory apparatus holds regardless of NY-12’s outcome. For AI companies evaluating political risk, the side prioritizing investment in public safety perception is better positioned in the political environment two-thirds skepticism creates, but only if disclosures about dark-money intermediaries do not weaken that positioning. For PAC strategists across both networks, the leading indicator that would trigger spending reallocation is a post-NY-12 poll showing voters in other targeted districts citing industry PAC spending as a salient issue rather than a background fact — this would convert the spending itself into a mobilizing signal for opponents, an outcome the 2024 crypto playbook did not face in an environment where the asset in question was not perceived as threatening livelihoods at scale.

Analytical techniques used in this piece

This analysis applies the methods below. Each links to a short, plain-English explainer you can read and reuse.

Frame Audit
Surfaces the frame an argument adopts and what that framing quietly includes or excludes.
Interest Mapping
Separates parties’ stated positions from their underlying interests (Fisher & Ury).
Scenario Planning
Builds a small set of distinct, plausible futures to plan against.