Public First PAC Midterms: 10 of 17 Wins After $63M AI Safety Spend

The 2026 US midterms closed a $100-million-plus brawl between two AI political networks. On one side, Leading the Future pushed lighter federal rules and more room for deployment. On the other, Public First backed candidates who want stronger guardrails and less self-regulation. Bloomberg reported on November 7 that Public First spent more than $63 million and that 10 of the 17 congressional candidates it backed won. Public First Action's own November 4 statement celebrates wins across both parties but does not print that win-rate line.

If you ship AI products in the United States, this is not abstract politics. The same election sent state AI bill authors to Congress and reinforced local fights over data centers. For the spending picture before election day, see our October FEC-based PAC roundup.

Update (November 8, 2026): how the safeguards coalition formed

Bloomberg's November 7 scoreboard tells you who won and how much cash moved. Politico Magazine's November 6 feature, relayed on Techmeme, explains how a small network became a counterweight to Leading the Future. We could not open the full Politico piece in this environment because of a paywall, but Politico's public excerpts and prior interviews give a usable timeline.

Politico dates the shift to July 2025, when a Senate vote on an AI preemption rider went 99–1 against the industry-backed language. Advocates who had treated federal AI law as a decade-long project suddenly faced a 16-month runway to the 2026 midterms. On the deployment side, Leading the Future launched in August 2025 with more than $100 million pledged, modeled on crypto-industry PAC spending, per Politico's summary and our October Leading the Future FEC roundup.

Politico describes safeguards donors and operatives coordinating in Signal group chats, then formalizing into the Public First network. Named architects in the piece include Brad Carson (Americans for Responsible Innovation and Public First co-founder), consumer attorney Jay Shooster, and Democratic strategist Rishi Bharwani. Carson told Politico the coalition "congealed over that fight, and it has sustained since then"—meaning the July preemption brawl, not election night.

That origin story matters for builders for two reasons. First, the money is not only Anthropic-branded charity; Carson has said Anthropic is the largest donor but not the strategist, a line consistent with Inside Political Money's October profile. Second, the same network explicitly framed tort liability as an underrated lever for AI governance in Carson's August Politico Q&A, which raises the odds that product-liability politics—not just model-release politics—stay on the 2027 agenda.

None of that changes the verified scoreboard lines in the table below. It does explain why Public First could answer a nine-figure rival within one cycle instead of waiting for a slow Senate framework bill.

TL;DR: scoreboard and sources

What Public First says happened

In its November 4 release, Public First Action, Jobs and Democracy PAC, and Defending Our Values PAC congratulate winners from California to Georgia. The post names candidates by district and ties many of them to AI oversight work:

• Scott Wiener in California's 11th, author of frontier AI bill SB 53, on track for Nancy Pelosi's former seat.
• Manny Rutinel in Colorado's 8th, sponsor of the Colorado Artificial Intelligence Act, defeating Republican incumbent Gabe Evans.
• Valerie Foushee in North Carolina's 4th, co-chair of the House Democratic Commission on AI, winning reelection.
• Brian Poindexter in Ohio's 7th and Bob Brooks in Pennsylvania's 7th, both boosted in crowded primaries.
• Republican winners include Celeste Maloy (Utah), Kevin Hern (Oklahoma Senate), Pete Ricketts (Nebraska Senate), and John Cowan (Georgia).

Semafor's November 5 recap stressed Democratic House wins for Public First Action in the general election and said Republicans the network backed lost. That count does not include several Defending Our Values PAC GOP wins the group listed on November 4.

Brad Carson, the group's Democratic co-founder, said: "Voters in both parties delivered the same verdict tonight" and called the night "a crushing blow to the AI industry's anti-regulatory efforts to buy a Congress that would look the other way." Republican co-founder Chris Stewart framed AI risks as bipartisan.

The release also says the three entities were active in at least 78 districts and states when you count polling, contributions, advocacy, and independent expenditures. That is a wider footprint than the 17-candidate Bloomberg endorsement list, so the two numbers measure different things.

AI super PAC midterm results shown as a ballot entering a secured ballot box

Where the $63 million and 10-of-17 figures come from

Bloomberg's Emily Birnbaum ties the $63 million figure and 10-of-17 win rate to Public First's own reporting. We could not open the full Bloomberg article in this environment because of a paywall, but the headline figures appear in Bloomberg's public summary and were relayed by Techmeme on November 7.

Until Public First or independent FEC tallies publish a single table that matches Bloomberg, readers should hold three facts at once:

1. Self-reported totals from 501(c)(4) arms can include money not yet spent and not fully visible in FEC independent-expenditure feeds.
2. Different arms (Jobs and Democracy, Defending Our Values, Public First Action) file on different schedules, which is why Semafor can cite $20 million for one arm in the general while Bloomberg cites $63 million for the network.
3. Win definitions differ. A group can "support" a candidate through a primary-only buy, a hybrid PAC, or issue ads that never mention AI.

Our earlier PAC post walked through the same counting problem in October, when safeguards groups looked narrowly ahead on FEC analyses through October 20.

How Leading the Future's night looked different

Wired's November 4 analysis frames Leading the Future as leaving the cycle with cash on hand and many safe-seat wins, but weak returns in contested races. The piece notes about $26 million in primary independent expenditures, then only about $10 million in the general, mostly on comfortable seats. Funding reportedly slowed, with zero donations reported in the second and third quarters.

Semafor quotes spokesperson Jesse Hunt saying the network is proud of "pro-innovation champions" who won and is committed through 2028. That matches Wired's read that Leading the Future shifted from intimidation-style primary fights toward ingratiating incoming members rather than swinging toss-ups.

The highest-profile loss on the deployment side was Darline Graham in South Carolina's Senate race. Semafor reported Leading the Future spent $4.8 million backing Graham; she lost to Democrat Annie Andrews, who campaigned with the nickname "Data Center Darline." Wired puts American Mission support above $2 million and lists the result at 46.5 percent in a state that last elected a Democratic senator in 1998.

Contrast that with Jared Moskowitz's Florida House race, where Think Big spending helped a Democrat hold a redrawn seat, per Semafor and Wired. The mixed scoreboard is why both camps claim partial victory.

Anthropic's money in the story

Anthropic Public First PAC funding framed as regulatory compliance documentation

Anthropic is not on the ballot, but its dollars are in every recap. Inside Political Money's October profile quotes Brad Carson saying Anthropic's commitments ultimately totaled $40 million, with the company stating that corporate funds were restricted to policy and public education rather than candidate election ads. Dario Amodei's $1 million personal gift to Public First PAC appears in the same piece.

Anthropic's wider 2026 political footprint, including its presidential engagement program for 2028, sits on a different track from super PAC independent expenditures. Carson has repeatedly insisted Public First does not take marching orders from Anthropic, pointing to his separate watchdog group Americans for Responsible Innovation, which criticizes labs including Anthropic.

Builders should not read "Anthropic won the midterms" from a 10-of-17 rate. They should read: a safeguards-aligned network had enough cash to answer Leading the Future's Fairshake-style threat, protect vulnerable incumbents, and send state-law authors to Washington.

Standout races that matter for AI law

Federal AI preemption debate as capitol columns beneath a policy flag

Several Public First-backed winners are policy writers, not generic partisans:

Scott Wiener and Manny Rutinel bring state-level AI statutes into the House. That matters because the industry fight over federal preemption of state AI laws is the central legislative lever. If preemption fails, builders keep tracking California, Colorado, and copycat states separately.

Valerie Foushee keeps a leadership slot on the House Democratic Commission on AI. Her narrow survival after heavy spending was already a story in the primary season covered by our data center backlash midterm piece.

Celeste Maloy, a Utah Republican, illustrates Public First's bipartisan arm. Defending Our Values backed her primary; she has put forward legislation examining nuclear-related risks from AI systems, per summaries relayed in secondary coverage.

The marquee loss for the safeguards camp remains Alex Bores in New York's 12th Democratic primary, where Leading the Future spent millions against the RAISE Act co-author and Public First spent heavily in reply. Bores lost, but Wired and Inside Political Money both treat the spend as a deterrent signal rather than a wasted check.

Data centers and the election AI voters actually saw

Money in Washington is only half the story. Semafor notes that about 40 local ballot measures curbing data center development passed on election night, while a GOP operative told the outlet Leading the Future did not resource fighting them. That aligns with our reporting on OpenAI's post-midterm messaging plans and the wider Democratic AI regulation ad fight.

Wired cites an NBC exit poll finding that 56 percent of voters said AI and data centers played a major role in their vote. Even if you treat exit polls cautiously, the combination of local moratoria wins and anti-data-center attack lines in Senate races shows voters experienced AI as an infrastructure and jobs question, not only a chatbot novelty.

What this means for what you build or pay

None of November's results instantly rewrite Claude, GPT, or Gemini terms of service. They change the probability distribution for 2027–2028 legislation:

If you run a US-facing product today, keep doing three boring things well: maintain a state AI law tracker, document model versions and data sources for downstream audits, and separate marketing claims from logged behavior before regulators ask.

Teams that bet everything on a single federal preemption bill should revisit that assumption. The midterm scoreboard suggests a Congress with more members who already wrote state rules, not a blank slate.

What is claimed vs verified

When numbers conflict, trust dated FEC independent-expenditure exports over press-release totals, and trust named race results over network-wide win rates.

What people are asking

Did the safeguards camp "beat" Leading the Future? They won the narrative battle in several contested races and sent more AI bill authors to Washington, but Leading the Future still backed many winners in safe seats. Wired's headline is blunt: the biggest pro-AI PAC network did not really move the needle in competitive general elections.

Will Congress pass AI law in 2027? Public First's Carson told Inside Political Money that larger debates have barely begun legislatively, and that Trump-era accelerationism still caps what bipartisan AI bills can do. The midterms changed the menu of sponsors; they did not pass a framework.

Does more regulation mean slower model releases for developers? Not automatically. It usually means more paperwork, evaluations, and procurement friction for enterprise buyers first, then consumer rules. Watch committee assignments in January, not PAC press releases.

Related reading

• AI super PACs spent about $78.5M through October
• Anthropic's 2028 presidential engagement program
• Data center backlash in 2026 midterm races
• Democrats, Tomorrow Together, and Leading the Future
• OpenAI and the reported "AI compute campus" rebrand
• Maine's data-center revolt in a Senate race
• After the blue wave: what midterms may change for AI rules

FEC totals, independent expenditure definitions, and race calls are accurate as of November 8, 2026; late filings can change dollar figures.