Last verified October 31, 2026, three days before the U.S. midterm elections. Campaign-finance numbers move daily; the FEC’s electronic filing system was reported offline on October 28, so the newest late-spending disclosures may not be public yet.
The popular version of this story goes: a few tech billionaires now out-muscle voters and governments. After tracing six documented episodes from the past eighteen months, my honest answer is that it’s half right, and the wrong half matters.
Elected governments have recently shown they can blacklist an AI lab, switch off two frontier models worldwide, and refuse (99–1) a federal ban on state AI laws. What AI executives hold is quieter and, in some ways, harder to check: money routed through opaque vehicles, the ability to set the terms of their own products, and a lobbying presence that no public-interest group matches.
Disclosure. I used Anthropic’s Claude to help research and draft this piece. Anthropic is a party in one case below (the Pentagon dispute), funds the group opposing the OpenAI-linked PAC in another, and filed for an IPO on October 26. OpenAI executives sit on the other side of the same fight. I have tried to hold both camps to the same standard, relied on court records, FEC data and named reporters, and marked every claim I could not verify at the source.
Key takeaways
- The most-hyped AI political machine mostly stalled. The OpenAI-linked super PAC network spent heavily in primaries, then went quiet before the general election, with only about 5% of its direct spending reaching contested general-election races, per The Intercept’s FEC analysis.
- The biggest AI-executive check of the cycle wasn’t about AI. Elon Musk’s $158 million into one Texas Senate race is roughly double what Puck estimates all AI-focused PACs have spent combined.
- Governments hold the heavier stick. The Pentagon designation, the June export directive, and the preemption executive order all came from elected officials, not CEOs.
- The real democratic gap is procedural. Decisions are being made by designation, directive, shell-company vendor, and private usage policy, rather than by statute and open debate.
- Voters trust neither side. In Pew’s February 2026 survey, 67% had little or no confidence in government to regulate AI, and 59% had little or no confidence in companies.
Contents
- What “power” actually means here
- Case 1: The primary that backfired (New York’s 12th)
- Case 2: The donor who dwarfs the AI PACs (Musk, Texas)
- Case 3: Congress said no, so everyone went to court
- Case 4: The CEO who said no to the Pentagon
- Case 5: The government pulls the plug
- Case 6: Brussels softens its own AI Act
- What the public thinks
- Scorecard and evidence ledger
- What the pattern adds up to
- Practical steps
- What would change my mind, and limits
- FAQ
- Sources
What “power” actually means here
A CEO cannot pass a law, command troops, or levy a tax. So “more power than elected officials” only makes sense if we ask a narrower question: where can a few private people shape public decisions faster than voters can react?
I tested four channels:
- Election money: super PACs and direct donations.
- The right to refuse: setting terms on how a government may use your product.
- Rule-writing influence: lobbying for or against the rules themselves.
- The counter-power: what governments do back, which any honest analysis has to include.
Case 1: The primary that backfired (New York’s 12th)
Leading the Future (LTF) launched in August 2025, backed by OpenAI president Greg Brockman, Andreessen Horowitz, and others. By its own account, it and its affiliates had raised over $125 million in commitments and held over $70 million entering 2026. Its first marquee target was Alex Bores, the Assembly member behind New York’s RAISE Act, who was running in the Democratic primary for the seat Jerry Nadler is leaving.
A rival network, Public First, launched months later. Anthropic pledged $20 million to its affiliated nonprofit, with a stipulation that the money not be used on elections, and Anthropic CEO Dario Amodei personally gave $1 million, according to The Intercept.
The result: the two networks put $21.3 million into that single race. Bores lost to Micah Lasher by about 4 points, per The Cornell Sun. Lasher had endorsements from Nadler and Gov. Hochul and at least $10 million in backing from Michael Bloomberg, which Transformer notes was more than LTF spent against Bores. The same reporting shows prediction-market odds for Bores rose from about 10% to 31% after LTF entered, and that he got a stream of favorable press.
So did the AI money decide it? Nobody can honestly say yes. Bores himself argues the opposition mostly elevated AI as an issue, according to The Intercept, which also notes many observers agree.
What happened next is the more interesting part.
Source: The Intercept analysis of FEC records, October 29, 2026. Figures run through October 26; the FEC filing outage may hide the latest spending.
LTF’s general-election spending was $7.4 million as of October 26 against roughly $34 million in cash across its network, and by The Intercept’s count, only 31% of that went to races rated competitive. Public First had spent $1.1 million on general elections. A separate San Francisco Standard report says Brockman pulled out of a second planned $25 million donation in June, calling the effort “a distraction,” and that adviser accounts say investor Ron Conway regrets his $500,000 gift. (Cash and spending totals differ between outlets because of filing dates and what counts as “the network.”)
One more finding deserves attention because it’s about transparency, not persuasion. The Intercept reports LTF’s affiliates routed more than $30 million through newly created Delaware corporations, including $21.3 million by Think Big and $17.4 million by American Mission, which a Campaign Legal Center lawyer told the outlet looks like a way to avoid naming real vendors. CLC filed an FEC complaint in May, but the FEC lacks a quorum to act. LTF has defended its practices as legal.
What this case shows: AI-industry money is real, large and, so far, politically toxic. A Data for Progress poll commissioned by Guardrails Alliance, a group fighting the AI PACs (read with that in mind), found about three-quarters of likely voters in Iowa, Michigan, and Ohio would be less likely to back a candidate supported by an AI super PAC. The danger isn’t that this money wins every race. It’s that voters in many races can’t see who paid for what.
Case 2: The donor who dwarfs the AI PACs (Musk, Texas)
The most literal “AI CEO with political power” in this cycle isn’t spending on AI policy at all.
According to FEC filings reported by CBS News and ABC News, Musk gave $75 million to the Senate Leadership Fund on September 9 and $83 million on September 23. Each time, the SLF sent an identical sum to a new affiliate, Texas PAC, a day later. Bloomberg’s reporting says Texas PAC had spent $148 million on the race between Ken Paxton and James Talarico by mid-October, and that Musk is the largest U.S. political donor for a second straight cycle.
Sources: Bloomberg via Business Standard; Puck’s FEC review via FourWeekMBA (secondary, treat as an estimate); The Intercept.
I want to be careful here. Musk’s money is aimed at a Senate seat, not at AI regulation, and his companies span much more than AI. But this is the point: if you’re asking whether AI executives can outspend the democratic process, the clearest example is one person, one race, one quarter. His reported $158 million alone is nearly twice the roughly $80 million Puck’s review puts on every AI-focused PAC combined this year.
What this case shows: The question “which AI company is buying politics?” misses the largest check, which belongs to an individual executive and flows through a conventional party vehicle. Whether you cheer or fear it depends on which candidate you like, and that’s exactly the problem for a rule-of-law approach.
Case 3: Congress said no, so everyone went to court
In July 2025, the Senate voted 99–1 to strip a ten-year ban on state AI laws from the budget bill. A second attempt through the defense bill also failed. On December 11, 2025, President Trump signed Executive Order 14365, creating a DOJ AI Litigation Task Force (formalized January 9, 2026) and, per Bloomberg Law, handing industry leaders a win. Travis Hall of the Center for Democracy & Technology had said beforehand that preemption is Congress’s decision.
Here is where the story gets concrete. xAI sued Colorado over its AI Act on April 9, 2026, and the DOJ moved to intervene, its first such step. According to one legal tracker, enforcement was suspended, and Colorado’s legislature then repealed and re-enacted a narrower law (SB 26-189) in May, effective January 1, 2027. That’s the closest thing in this study to industry pressure producing a policy change through ordinary government channels, though I’ve only confirmed the repeal through a secondary source, so check Colorado’s legislative record before citing it.
At the federal level, the same tracker (Astraea Law, October 27) says no state AI law has been preempted, no general federal AI statute exists, and the Commerce Department’s required list of “onerous” state laws still hadn’t been published.
What this case shows: The preemption push has produced pressure, lawsuits, and rewrites, but not the legal override its backers wanted. The democratic point cuts both ways. An elected Senate rejected preemption by a landslide, and an elected president pursued it anyway by other means.
Case 4: The CEO who said no to the Pentagon
Anthropic, my tool’s maker, is the company in this case. I’ve kept to court records and wire reporting.
Anthropic refused to let its models be used for fully autonomous weapons or mass domestic surveillance of Americans. Defense Secretary Pete Hegseth had warned it would lose a $200 million contract or be designated a supply-chain risk, and in February President Trump ordered agencies to stop using its products. A March 3 letter made the designation official. It was the first time a U.S. company had been publicly designated under that kind of statute, which was written to counter foreign sabotage.
The courts have split. In California, Judge Rita Lin first blocked the move in March and in late August ruled the Pentagon had violated Anthropic’s First Amendment rights, a ruling that blocks the broader federal ban. But on September 25, the D.C. Circuit, 2–1, upheld the Pentagon’s separate designation. Judge Katsas’s majority accepted the Pentagon’s reasoning that a model built to refuse some uses is an operational risk, and Judge Henderson dissented. Anthropic asked for en banc rehearing on October 13, and per reporting on October 27, the DOJ filed a notice of appeal to the Ninth Circuit in the California case. This is live litigation, and the situation may have changed since I checked.
Give the Pentagon its best argument: a military cannot depend on software a vendor could stop it from using mid-operation. Anthropic’s best argument is that the law was built for foreign threats and now lets the government punish a domestic vendor for its stated safety limits.
What this case shows: A private company set an ethical boundary, and the government responded with coercion. Both sides were making what is really a policy decision about autonomous weapons and domestic surveillance. Neither was elected to make it, and Congress hasn’t.
Case 5: The government pulls the plug
On June 12, 2026, the Commerce Department ordered Anthropic to block foreign nationals from its new Claude Fable 5 and Mythos 5 models. Because the company couldn’t verify nationality in real time, it disabled both for everyone. Controls were lifted June 30 after Anthropic accepted conditions Commerce Secretary Howard Lutnick set, and access returned July 1. Anthropic’s own IPO risk disclosures now cite this episode.
What this case shows: The cleanest counterweight to the headline. A flagship AI product vanished worldwide for roughly two and a half weeks on a regulator’s directive. If you build anything on a single frontier model, this is the case to remember.
Case 6: Brussels softens its own AI Act
Corporate Europe Observatory and LobbyControl found digital-industry lobbying in Brussels rose from €113 million in 2023 to €151 million, a 33.6% jump. They also found that seven of eight Digital Omnibus changes mirrored industry positions, and that 69% of Commission meetings in 2025 were with business groups versus 16% with NGOs. EU governments and lawmakers then agreed to push high-risk AI obligations from August 2, 2026 to December 2, 2027.
| Indicator | Figure | Source |
|---|---|---|
| Digital-industry lobbying spend, 2023 | €113M | CEO / LobbyControl |
| Digital-industry lobbying spend, 2025 | €151M (+33.6%) | CEO / LobbyControl |
| Omnibus changes mirroring industry positions | 7 of 8 | CEO / LobbyControl |
| Commission meetings with business vs. NGOs (2025) | 69% vs. 16% | CEO / LobbyControl |
| High-risk AI deadline | Aug 2, 2026 → Dec 2, 2027 | Reuters via BHRRC |
Counterpoint: supporters of the delay point out technical standards weren’t ready, and the Commission cited pressure from Washington as well as industry. Max Schrems of noyb, a privacy group, argues the package mostly benefits Big Tech rather than ordinary European companies.
What this case shows: Lobbying is legal and routine. The signal is the access gap, and the fact that outside pressure from a government (the U.S.) and from industry pushed in the same direction at once.
What the public thinks
Pew’s June 2026 report (survey of 5,119 adults, February 17–23) found Americans using AI more and trusting the referees less.
Source: Pew Research Center. Chatbot use was 33% in 2024.
One shift matters politically: Democrats now doubt the government’s ability to regulate AI more than Republicans do (74% vs. 61%), a reversal from 2024, per Pew. The public wants rules and doesn’t believe anyone will write good ones. That vacuum is where private actors fill in.
Scorecard and evidence ledger
| Case | Channel | Who held the stronger hand | Evidence level |
|---|---|---|---|
| NY-12 primary | Election money | Neither; result inconclusive | FEC-based reporting; causal impact unproven |
| Musk / Texas Senate | Individual donor | The donor (in money terms) | FEC filings, multiple outlets |
| State-law preemption | Lobbying + executive action | Contested; no preemption yet | EO text verified; legal tracker for status |
| Pentagon dispute | Right to refuse | Government so far; courts split, ongoing | Court rulings, wire reporting |
| Export controls | Regulatory power | Government | Multiple outlets, company statement |
| EU Omnibus | Rule-writing | Industry plus U.S. pressure | NGO analysis; motives disputed |
Evidence confidence at a glance: Strong (court rulings, FEC filings reported by several outlets): Pentagon rulings, Musk donations, export controls. Moderate (single analytic source): AI-PAC total spending, Delaware-vendor pattern, EU lobbying analysis. Weak or contested: any claim that AI money decided a specific race.
What the pattern adds up to
If I had to name it, I’d say AI executives rarely overrule elected officials. They shape what reaches the ballot and the bill, how fast, and who can see it happen. Meanwhile, elected officials increasingly use fast executive tools (designations, directives, orders) that skip the slow public process.
Three things stand out:
- Money is loud but leaky. The most explicit AI-industry political spend backfired in public and retreated. The quieter version, such as vendor shells and personal mega-donations, is harder to track.
- The stick is asymmetric. A company can refuse. A government can respond by blacklisting or switching off. Courts are still deciding where the line sits.
- Nobody voted on the underlying questions. Whether AI should run weapons, whether it should surveil citizens, whether a state can regulate it: those decisions are being improvised in procurement disputes and lawsuits, while Congress has not legislated.
“Which billionaire do you trust?” is the wrong question. “Who decided, and could I see it happen?” is the better one.
Practical steps
If you’re a voter
- Check any race with heavy outside ads in the FEC’s independent expenditures data. Look up who funds the PAC, not just its name, and note that late filings may lag while the e-filing system is disrupted.
- Treat ads that never mention AI with caution: much of the early AI-PAC advertising was about other issues.
If you’re a journalist or publisher
- Separate three things in every story: money spent, outcome, and proof the first caused the second. NY-12 shows how easily the third gets assumed.
- Label evidence strength (filed record, single-source report, estimate) as I did in the ledger above.
If you’re a policymaker
- Put the big calls into statute: autonomous-weapons limits, domestic-surveillance limits, and when the government may cut off a domestic vendor.
- Require vendor-level disclosure for super PAC spending, so shell-company routing stops being a workaround.
If you run or work in an AI company
- Publish your usage red lines and your political spending in one place. Disputes like the Pentagon case will keep arriving, and the public currently can’t evaluate them.
If you build on frontier models
- Keep a tested fallback provider. The June shutdown proved a model can disappear overnight for reasons outside the vendor’s control.
What would change my mind, and limits
I’d revise the “half right” verdict if: late-October filings show a large, undisclosed AI-PAC push into contested races; the Ninth Circuit or en banc D.C. Circuit decisions broaden or narrow the government’s power over vendors; or Colorado-style repeals spread to other states under litigation pressure.
Limits: this is a U.S.-heavy study with one EU case. Dollar figures come from press analyses of FEC filings and may shift. The Pentagon litigation was moving as of October 31. Some statements (Colorado’s repeal, the Commerce list status) rest on legal trackers rather than primary records, and I’ve flagged them. Nothing here shows that AI money decided any election.
FAQ
Do AI CEOs really have more power than elected officials? No, not in the sense of legal authority. Elected governments recently blacklisted an AI lab, switched off two models, and rejected a federal ban on state AI laws. What AI executives hold is influence over timing, money, and product terms, often with less transparency.
How much have AI super PACs spent in the 2026 midterms? Estimates vary by how groups are counted. The two main networks put $50.7 million into primaries according to The Intercept, and Puck’s review puts all AI-focused PACs near $80 million. Check FEC data for current totals.
Did AI money defeat Alex Bores? Unproven. Bores lost by about 4 points; Michael Bloomberg’s $10 million-plus for the winner exceeded what Leading the Future spent against him, and several observers say the opposition raised Bores’s profile.
Is the Pentagon case over? No. The D.C. Circuit upheld the Pentagon’s designation 2–1 on September 25; Anthropic has asked for en banc review, and the DOJ has appealed the California ruling to the Ninth Circuit.
Sources
Campaign finance and politics
- The Intercept, Oct. 29, 2026: AI super PAC spending analysis
- San Francisco Standard, Oct. 31, 2026: AI industry super PAC retreats
- Transformer: What Alex Bores’ defeat tells us
- FEC: Leading the Future committee page
- CBS News: Musk’s Senate Leadership Fund donations
- Public Citizen: Corporate spending in the 2026 midterms
Law and courts 7. NPR, Aug. 28, 2026: Judge rules against Pentagon designation 8. Law Commentary, Sept. 25, 2026: D.C. Circuit ruling 9. Bloomberg Law: Executive Order 14365 10. Jenner & Block: DOJ joins xAI v. Colorado 11. The Next Web: Export controls lifted
Europe and public opinion 12. Tech Policy Press: Corporate Europe Observatory analysis 13. EU Perspectives: Interview with Bram Vranken (CEO) 14. Pew Research Center: Americans and AI 2026