After the November 2026 U.S. midterm elections, a bill was placed on the technology industry’s table: Public First Action, the AI safety faction PAC whose largest single donor was Anthropic, spent $63 million, and 10 of the 17 congressional candidates it backed won. Leading the Future, the accelerationist PAC supported chiefly by Andreessen Horowitz and OpenAI co-founder Greg Brockman, along with its allies, invested more than $48 million in total, with comparatively weaker results. Combined, the two factions channeled more than $110 million in AI political funding into the 2026 midterms. According to multiple media reports, this was the first time in U.S. history that a stance on AI regulation became a core issue in congressional campaigns.
To understand the real logic of this contest, one must first clarify where it began. In February 2026, according to CNBC, Anthropic donated more than $20 million to Public First Action, the largest single political donation the organization had received since its founding. Public First Action is jointly led by former members of Congress Brad Carson and Chris Stewart and is explicitly positioned as a bipartisan force supporting AI regulation. It planned to support 30 to 50 candidates in this election cycle, with a target fundraising range of $50 million to $75 million. In the end, actual spending closed at $63 million, above the upper end of the original range. The money was not scattered at random—according to Semafor, Public First’s direct investment during the general election phase was $20 million, while an affiliated organization called Guardrails Alliance spent an additional $3.1 million on opposition ads against five candidates, all of whom ultimately lost.
The funding sources on the opposing side are equally worth examining. Leading the Future initially entered with great fanfare and a war chest of more than $100 million. Its backers, besides Andreessen Horowitz, included Palantir co-founder Joe Lonsdale, SV Angel founder Ron Conway, and AI search company Perplexity. The position this camp represents is that the federal government should clear regulatory obstacles for AI technology development and promote a unified national framework that can override state regulations—indeed, this is also the core goal the accelerationists hope to achieve through legislation in Congress: using a federal framework to preempt regulatory rules already implemented in the states.
The two factions’ choice of battlegrounds reveals their respective strategic intentions. The safety faction’s precision was concentrated in three swing states: in Colorado, according to multiple media reports, Public First invested $1.5 million in a PAC supporting Manny Rutinel, and the candidate, who had advocated a “conditional pause” on AI development, ultimately won; in Wisconsin, safety faction groups spent more than $880,000 on ads against rival candidate Derrick Van Orden, and the candidate they backed, Rebecca Cooke, won the seat there; Michigan also flipped successfully. The accelerationists paid their most expensive price in the South Carolina Senate race—according to TechCrunch, Leading the Future poured $4.8 million into the campaign of Republican candidate Darline Graham, but she ultimately lost to Annie Andrews. During the campaign, Andrews branded Graham as “Data Center Darline,” with the harm that data center construction does to local communities as her core attack, and the label had a clearly damaging effect among voters. In Montana, Aaron Flint also lost after receiving more than $500,000 in support from the accelerationists.
From this battlefield layout, a structural judgment can be drawn: what the safety faction won was not only seats, but the continuation of their identity as legislative agents. Among the winning candidates, several already had practical legislative experience—Manny Rutinel was a key driver of Colorado’s comprehensive AI regulation law, and Scott Wiener led the passage of multiple AI transparency and safety laws in California. Their entry into Congress means that first-hand experience with state-level regulation will be brought directly into the federal legislative debate. This is policy penetration of a completely different order of magnitude from merely using donations to influence a candidate who is not familiar with AI.
The accelerationists’ dilemma came not only from their opponents’ ability to spend, but also from a deeper structural headwind: public sentiment. According to multiple surveys, about three-quarters of likely voters said that upon learning a candidate had accepted money from an AI super PAC, they would be less likely to vote for that candidate. This means that in some competitive districts, accelerationist money itself had become a burden. The South Carolina case is a typical positive feedback loop: the more that was spent, the more was lost, as opponents turned the funding source itself into a campaign issue. At the same time, about 40 local ballot measures restricting data center construction passed across the United States in this election. According to Semafor, these grassroots backlashes were almost impossible for any AI-backed PAC to effectively intervene in—capital-intensive lobbying still has room to operate in national elections, but it is highly inefficient in the face of local direct democracy.
Looking at the overall election results, the Democrats won control of Congress in this midterm election. This broader backdrop, combined with the AI safety faction’s victories, formed a political combination relatively favorable to regulatory legislation. Against the backdrop of the general election, the Democratic Party as a whole holds a more proactive attitude toward AI regulation, and the safety faction’s candidates within the party had previously received considerable support resources.
But an important distinction must be made here: between election victory and legislation becoming law, there is a wide zone of uncertainty. The safety faction won 10 federal seats, which in a Congress of hundreds of seats is still not enough to push any legislation on its own—real legislative progress depends on the actual degree of bipartisan cooperation and on whether these newly elected members can secure time and resources for AI regulation in the competition over Congress’s priority agenda. According to reports, the details of specific legislative proposals and the depth of bipartisan cooperation remain to be seen. After the election results were announced, a spokesperson for Leading the Future said the organization would view this election as the starting point of a “multi-election-cycle effort” and pledged to focus on laying groundwork for 2028. This means the accelerationists do not intend to retreat from the political contest, but rather to extend the battle lines.
From the perspective of engineering practice, the significance of this political contest for AI developers and enterprise users is not abstract. The core disputes over regulatory legislation center on two levels: first, whether a federal framework takes precedence over state regulations—if the accelerationists’ logic of “federal override of states” is pushed in the next Congress, then the specific compliance requirements already in effect in Colorado and California will face the risk of being overridden; conversely, if the federal framework pushed by the safety faction uses existing state-level regulations as a baseline, compliance costs will only rise. Second, there are deployment restrictions on high-risk AI systems—the “conditional pause” stance supported by Rutinel and others means that in some scenarios, deployment restrictions or mandatory assessment requirements for certain categories of AI systems may emerge at the federal level.
At this current juncture, the true width of the 2027 legislative window is roughly as follows: Democratic control of Congress, a net increase in the number of safety faction members, and the entry into the federal legislature of direct carriers of Colorado and California legislative experience—these three factors combined move some form of federal AI regulatory framework legislation from possibility to higher probability. But the specific shape of this framework, whether it will be mandatory or voluntary, and whether it will replace state laws with federal standards or layer federal standards on top of them, remain unclear. Specific indicators that can be observed as signals include: the committee assignments of Manny Rutinel and Scott Wiener in Congress (whether they enter technology- or commerce-related committees), and the 2028 funding target amount announced by Leading the Future—the latter will reveal the accelerationists’ true assessment of this defeat.
This $110 million midterm battle ultimately presents a pattern that is not rare in political history but is clearly visible for the first time in AI policy history: the internal route divisions within the technology industry have completed their migration from intra-industry debate to electoral politics. In this sense, what happened in November 2026 was not merely an election result, but a confirmation moment that the AI regulation dispute has formally entered the dimension of sovereign contest. The above analysis is based on existing public information and is a judgment rather than established fact.
Sources: - [The Winners and Losers of AI's First Big Election](https://time.com/article/2026/11/04/midterm-elections-results-ai-industry-data-centers/) - [AI PACs spent millions trying to shape Congress, here's how they fared](https://techcrunch.com/2026/11/04/ai-pacs-spent-millions-trying-to-shape-congress-heres-how-they-fared/) - [Where AI money worked in the midterms](https://www.semafor.com/article/11/05/2026/where-ai-money-worked-in-the-midterms) - [Anthropic gives $20 million to group pushing for AI regulations](https://www.cnbc.com/2026/02/12/anthropic-gives-20-million-to-group-pushing-for-ai-regulations-.html) - [AI safety advocates make gains after $110 million midterm fight](https://www.japantimes.co.jp/business/2026/11/08/tech/ai-safety-us-midterms/)© 2026 Winzheng.com 赢政天下 | 转载请注明来源并附原文链接