Many people support stricter regulations on frontier AI, and some of those people donate to political campaigns for pro-regulation candidates. Not many people oppose AI regulations, but that group includes AI companies and execs who have giant piles of cash to throw at super PACs.

This situation is made possible by three court decisions:

  1. Buckley v. Valeo (1976) held that limits on independent political spending violate the First Amendment (while limits on direct contributions to candidates do not).
  2. Citizens United (2010) extended those First Amendment protections to corporations and unions, thus allowing them to spend unlimited money on political ads.
  3. SpeechNow.org v. FEC (2010) built on Citizens United to allow unlimited donations to independent-expenditure-only committees—i.e., super PACs—as long as they don’t directly work with political candidates.

Donations straight to a candidate are still capped at a few thousand dollars per person; donations to a super PAC supporting that candidate are not. If SpeechNow were overturned, or a law passed overriding it, the $5,000 cap on super PAC contributions would apply again, which means the popularity of a position would matter more than how rich its supporters are.

(I often hear “Overturn Citizens United” as the slogan, but SpeechNow allows super PACs to exist, and Buckley v. Valeo enables unlimited spending by individuals. What the advocacy groups in the space really want to do is restore the power of Congress and the states to enforce spending limits.)

Restoring spending limits would prevent unpopular but rich anti-regulation special interests (like Leading the Future) from dominating the political funding landscape. This would make the AI regulation situation better, but it would also improve the situation on any policy issues where unpopular, wealthy special interests wield disproportionate (and arguably un-democratic) influence.

It would also worsen any situation in which rich special interests are in the right, and the majority is in the wrong. But there aren’t many of those, and anyway there’s good reason to dislike an un-democratic process even if it produces a good outcome in a particular case.

An important caveat: Before writing this post, I was confident that imposing campaign spending limits would be a good thing. After doing some more research, I find it hard to dismiss the free speech defense of unlimited campaign spending (I’ll say more about this in the next section). Both sides of the debate have points in their favor. My thesis in this post should be taken as the narrow claim that campaign spending limits would have positive first-order effects on AI extinction risk, and they’re worth exploring on that basis. I don’t have an established view on whether stricter spending limits would be good or bad all-things-considered.

Contents

In defense of unlimited campaign spending

The typical argument in support of these decisions is “campaign spending is speech.” I never understood what that meant until I started writing this post. It seemed silly to me before, but it no longer seems silly. Here’s how I would explain it to my past self:

Say I really like peaches. Every time I meet a new person, I tell them about how much I like peaches. I decide I need to reach even more people, to spread the good word of peaches. I start buying TV ads about peaches. A strange decision? Maybe. A waste of money? Arguably. But I’m exercising my constitutionally-protected right to free speech, and the government has no right to tell me not to run those ads.

Now say I really like a political candidate. I would like to tell everyone that I like the candidate. At minimum, we can agree that the government can’t tell me not to tell all my friends about them—that would be a clear violation of free speech.

Maybe I want to buy ads telling everyone that I like that candidate. If I’m allowed tell all my friends about the candidate, or I’m allowed to buy ads telling everyone how much I like peaches, then why would I not be allowed to buy ads telling everyone how much I like a political candidate?

The opposite argument is that allowing unlimited campaign spending gives disproportionate influence to the wealthy, which violates the basic democratic principle of “one person = one vote”.

I do not have a strong belief on whether advertising for a political candidate is protected by the First Amendment—I can see the case either way.

In the rest of this post, I will discuss what would be involved in restoring spending limits, with the caveat that I’m not sure that would be consistent with preserving constitutional rights, and I’m not sure how to weigh that against the fact that doing so would decrease the chances that we are all killed by superintelligent AI.

Who’s working on this already?

The issue of wealth inequality in campaign spending already receives significant attention—I frequently hear people complaining about Citizens United. The amount of complaining is disproportionate to the amount of actual work being done, but there are a few advocacy groups working to restore spending limits. (See Appendix A for a table of orgs and their budgets.) Based on public tax filings (compiled by Claude), those orgs spend on the order of $10–15 million/year on the issue.

That’s not a small amount of money, but it’s small enough that a large foundation could comfortably double the size of the field.

Advocates have tried several strategies, including:

  • Pass a new law or even a constitutional amendment (e.g., Abolish Super PACs Act or Citizens Over Corporations Amendment). These tend to fall along party lines, with Democrats supporting but Republicans opposing. People have been trying for decades with no success.
  • Litigation to put SpeechNow in front of the Supreme Court, which has never reviewed it. In 2016, the advocacy group Free Speech for People filed a lawsuit against the Federal Election Commission challenging the existence of super PACs. In 2020, the Supreme Court declined to review the case. (This case is notable for the fact that the plaintiffs were bipartisan members of Congress—as opposed to legislation proposals, which have been heavily Democrat-leaning.)
  • “The Montana Plan” (Transparent Election Initiative): State-level laws could impose restrictions on election spending. This is a new plan—a ballot initiative will appear on Montana’s 2026 ballot. The initiative would regulate corporations through state government, although it does not restrict individuals’ campaign spending.

Would it be cost-effective to try to restore spending limits?

Probably not?

Legislation is unlikely to make it through Congress. People have been trying since 2010—the DISCLOSE Act, for example, was introduced and failed to pass, and has been re-introduced several times since with little progress. However, Lieu v. FEC (mentioned in the previous section) had bipartisan support, so there may be some flavor of campaign finance reform that could get enough momentum to become law.

The intractability alone is not necessarily defeating, because there is no clearly-tractable strategy for preventing AI from killing everyone.

If we’re specifically talking about AI extinction risk (not promoting democracy in general), then another problem is the significant lag time. There’s a long chain between “advocate for campaign spending limits” and “AI doesn’t kill everyone”:

  • Step 1. You support advocacy groups.
  • Step 2. Some time later, a law is introduced, a lawsuit is filed, etc.
  • Step 3. In the best-case scenario where the first attempt is successful, spending limits are restored one or a few years later.
  • Step 4. Wealthy anti-AI-safety interests can’t spend as much money on the next election cycle.
  • Step 5. The next election cycle happens; pro-AI-safety candidates perform better than they otherwise would have.
  • Step 6. Those pro-safety candidates eventually do things that reduce extinction risk.

The initial advocacy efforts don’t pay off until 2030 or 2032 at the earliest, at which point we may be past the point of no return.

In conclusion:

  • Advocating to restore limits on election spending may or may not be justified on general principle. (I am a strong supporter of free speech, but it seems ambiguous to me whether campaign spending qualifies as speech.)
  • On pure extinction risk grounds, restoring spending limits would be good, but efforts are unlikely to succeed. And even if they did, there’s a long lag time before advocacy efforts pay off.

Appendix A: Organizations working on restoring spending limits

I asked Claude Opus 5 to compile a list of relevant organizations. These are the ones that (according to Claude) spend at least 25% of their budgets on this issue. Some larger orgs have also done work on this as a small part of their activities; those include Brennan Center for Justice, Common Cause, and Public Citizen (Democracy Is For People).

Data is taken from tax filings on ProPublica and the FEC; I did not verify that the figures are correct. Claude came up with the figures for “Estimated share of spending” based on reading the orgs’ activities as qualitatively described in their tax filings; they are not based on hard data.

Total expenditures are estimated at $10–15 million per year, including the orgs below plus the larger ones. That includes perhaps 40 full-time employees (based on reported payroll expenditures, assuming an all-in cost per employee of $200,000) plus some unknown number of volunteers.

Organization Expenses (FY) Est. share of spending Source Notes
American Promise, Inc. (501c4) $3,163,099 (FY2024) 60–100% ProPublica, EIN 47-4608840 Direct read of extracted 990 data
American Promise Education Fund (501c3) $4,008,114 (FY2024) 60–100% ProPublica, EIN 47-4601462 Direct
Free Speech for People $2,347,309 (FY2024) ~25% (guess) ProPublica, EIN 45-0709993 Expenses direct; share is a guess based on campaign spending limits being roughly one of four program areas
Move to Amend $268,204 (FY2024) ~100% ProPublica, EIN 46-4306740 Direct
Wolf-PAC (federal committee) $234,424 (2023–24 cycle) ~100% FEC, C00485102 Two-year cycle, so ~$117k/yr

(Note: American Promise’s two entities are an affiliated c3/c4 pair; some money may be double-counted between them.)