What Election Law Was Written For and What It Is Now Facing

The Federal Election Campaign Act of 1971, and its significant amendment in 1974 following Watergate, established the basic architecture of American campaign finance law. The core mechanism: disclosure. Campaigns must report who donates money and how they spend it. Political advertising must identify who paid for it. The system assumed that political persuasion required spending money, that spending money left a paper trail, and that public visibility into that trail would enable accountability.

These assumptions were defensible in 1971. Television advertising was the dominant medium for political persuasion, and television advertising required buying airtime, which required a transaction, which left a record. The cost structure of political advertising was high enough that only funded campaigns and organized interest groups could afford significant media presence. The disclosure framework was designed for a world where political influence required money and where money was legible.

Every assumption underlying that framework has now been broken.

The Disclosure Requirement and What It Cannot Catch

The FEC’s disclosure requirements apply to “electioneering communications” — defined as broadcast, cable, or satellite communications that refer to a clearly identified candidate within sixty days of a general election or thirty days of a primary. The definition was written in an era when broadcast television and radio were the relevant media.

YouTube, Meta, TikTok, X, Instagram, Snapchat, and the rest of the digital advertising ecosystem are not broadcast, cable, or satellite communications in the FEC’s statutory definition. The FEC has attempted, imperfectly, to extend its disclosure requirements to digital political advertising through regulatory interpretation. Those extensions are contested, inconsistently applied, and do not address the fundamental issue: disclosure requirements assume a paid transaction, a purchaser, and an identifiable publisher.

AI-generated political content can circulate entirely outside any advertising transaction. A synthetic video shared organically through social networks has no “paid for by” disclosure, because no one paid for its distribution — only for its creation, and creation costs have fallen to the point where they are irrelevant to the campaigns with serious resources. When a political advocacy group creates 10,000 AI-generated social media posts and operates 500 AI-managed personas to distribute them, there is no “advertisement.” There is no media buy. There is no disclosure trigger. There is, at most, the cost of computing infrastructure — and that may be deductible as a business expense in ways that make it nearly invisible to campaign finance reporting.

Related reading

The Coordination Problem Becomes Insoluble

Campaign finance law draws a sharp distinction between “coordinated” expenditures, which count against contribution limits and require disclosure, and “independent” expenditures, which do not. The theory is that independent advocacy — organizations spending money without direction from a campaign — is protected speech, while coordinated spending is effectively a campaign contribution and should be limited and disclosed.

The practical application of this distinction has always been difficult, because coordination is hard to prove and campaigns have become sophisticated at operating through technically-independent structures that share strategic priorities. The FEC’s coordination rules require demonstrating either direct communication between the campaign and the outside group or use of campaign materials, plans, or strategy.

AI makes this distinction meaningfully harder to apply. Consider: a campaign develops a strategic messaging framework emphasizing border security. An independent super PAC develops an AI system trained on public campaign materials and strategically aligned political content. The AI system, trained on public inputs, independently generates content that closely mirrors the campaign’s messaging, because it was trained on the same strategic sources. No direct communication occurred. No campaign materials were shared. The coordination rules were not technically violated. The practical effect is indistinguishable from coordination.

This is not a theoretical future problem. Political consultants have described to reporters in 2026 the use of AI systems to produce “strategic alignment” between campaigns and nominally independent expenditure groups through parallel AI training rather than direct communication. The legal gray zone is wide and actively exploited.

State Law Is More Varied and Not More Helpful

The patchwork of state laws governing AI in political advertising is sometimes presented as a silver lining — states are experimenting, developing approaches that might be adopted federally. The reality is more complicated.

California’s AB 2655, enacted in 2024, requires platforms to label AI-generated content in political advertising during the 90 days before an election. Texas’s HB 4337 prohibits “deepfake” content in political advertising without the depicted person’s consent. Minnesota, Washington, and Georgia have passed various disclosure and prohibition measures.

The problems with this patchwork are multiple. First, enforcement. State laws regulating digital content face a structural problem: the content is not distributed through state-specific channels. A deepfake video produced in Florida, hosted on servers in Ireland, and distributed by an organization incorporated in Delaware circulates to California voters through a platform operating under federal rather than state jurisdiction. California can regulate what California-based platforms do. It cannot regulate what happens to content before it reaches those platforms or how it circulates through channels California doesn’t control.

Second, definitional inconsistency. “AI-generated” is not a legally precise category in any of these state statutes. A video shot on a real camera, edited using AI tools to modify specific statements, re-encoded and compressed until the modifications are not obvious — is that “AI-generated content” under California’s statute? The statute doesn’t say, and neither do the regulations implementing it. Campaigns and political operatives have found ways to operate in the definitional gaps.

Third, the federal preemption question. Several of these state laws are facing First Amendment challenges, and some are likely to be partially or fully preempted by federal law in contexts where federal jurisdiction is clear. The legal uncertainty reduces deterrent effect; if the law might be unconstitutional, the rational actor treats it as less binding.

The First Amendment Complication

American campaign finance law operates under a constitutional constraint that most other democracies’ equivalent laws do not face in the same form: the First Amendment’s protection of speech, particularly political speech, is extraordinarily strong under Supreme Court interpretation.

Buckley v. Valeo (1976) held that limits on campaign expenditures were unconstitutional restrictions on speech. Citizens United v. FEC (2010) extended this to corporate political spending. These decisions reflect a particular First Amendment theory: that political speech is so central to democratic self-governance that restrictions on it require compelling justification, and that expenditure limits — as opposed to disclosure requirements — generally don’t meet that bar.

This constitutional landscape significantly limits what Congress can do about AI-generated political content through campaign finance law. A prohibition on AI-generated political advertising would face a near-certain First Amendment challenge and would likely fail under current doctrine — it is a restriction on the content of political speech based on how it was produced, and that kind of content-based restriction faces strict scrutiny that it is very unlikely to survive.

Related reading

Disclosure requirements — requiring labeling of AI-generated content — are more constitutionally defensible, because Citizens United explicitly endorsed disclosure requirements while striking down expenditure limits. But disclosure requirements, as noted above, only work when there is a disclosable transaction, an identified speaker, and a platform obligation to carry the disclosure.

When the content is distributed organically through fake personas with no paid advertising involved, there is nothing for a disclosure requirement to attach to.

What Other Democracies Are Trying

The European Union’s approach, through the Digital Services Act and the proposed revisions to the Political Advertising Regulation, is structurally different from the American approach in ways that matter.

The EU framework puts obligations on platforms rather than (only) on advertisers. Large platforms — those with over 45 million monthly active users in the EU — are required to maintain political advertising transparency repositories, implement systems to detect and label AI-generated content, and conduct risk assessments for how their algorithms might affect political discourse. The platform is the regulated entity, not just the advertiser.

This is more promising than the American advertiser-focused approach because platforms actually have the technical capacity to implement these measures. A campaign or a bad actor can generate AI content and distribute it through organic means that evade advertiser-focused disclosure requirements. But the platform through which the content travels has, in principle, the technical capacity to detect certain patterns of inauthentic coordination and to label AI-generated content.

In practice, EU platform regulation is also not working as well as hoped. Enforcement of the DSA’s political content provisions has been slow and inconsistent. Meta’s compliance with the AI labeling requirements has been described by EU regulators as technically compliant but functionally inadequate — the labels are present but not prominent, and internal testing by researchers suggests most users don’t register them.

The UK took a different approach with its Online Safety Act, focusing on harm categories and imposing duties of care on platforms rather than specific political content rules. The UK’s approach has produced enormous legal and operational complexity for platforms and has been criticized both for overreach and for not adequately addressing political manipulation specifically.

None of these approaches has solved the problem. They have produced regulatory overhead, some deterrent effect at the margins, and ongoing political controversy — while the underlying technology has continued to advance faster than any regulatory framework.

The Enforcement Gap That Cannot Be Closed by Better Rules

There is a deeper problem beneath the specific legal inadequacies, and it is this: effective enforcement of any regulation governing AI-generated political content requires attributing that content to someone.

For a disclosure requirement to work, you must be able to identify who created and distributed the content. For a prohibition to work, you must be able to identify who violated it. For a platform removal obligation to work, the platform must be able to identify what to remove.

AI-generated content distributed through multiple layers of anonymization — created by an AI system, uploaded through a VPN, distributed through AI-managed personas with no connection to any real identity — is extraordinarily difficult to attribute. The attribution problem is not just a technical challenge (though it is that too). It is a structural feature of how the internet was designed: the architecture prioritizes routing efficiency over identity verification, and years of legitimate privacy advocacy have made strong anonymous communication an expected feature of the network.

A world in which political deception is easily attributable would also be a world with much less privacy protection for legitimate anonymous speech — whistleblowers, dissidents, journalists protecting sources. The privacy protections that make AI disinformation hard to attribute are the same privacy protections that protect genuine democratic speech. You cannot engineer away the anonymity problem without engineering away the privacy that democratic society also depends on.

Related reading

What an Honest Assessment Concludes

The gap between what election law says and what AI tools now enable is not a gap that better drafting will close. It is a gap between a legal framework built for one technological reality and a new technological reality that changes the fundamental parameters.

The 1971 campaign finance framework assumed that influence required money, that money left records, and that records enabled accountability. None of those assumptions hold for AI-generated organic political content. The disclosure and transparency framework built on those assumptions can be patched, extended, and improved at the margins. It cannot be made adequate to a world where creating and distributing political content at scale requires essentially no money, leaves no mandatory paper trail, and can be attributed to no one.

That doesn’t mean regulation is useless. Platform obligations, criminal penalties for specific harmful uses (synthetic endorsements of real people without consent, deepfake pornographic political targeting, which has been documented), and civil liability for provably false political deception all have marginal deterrent value. They will not transform the information environment.

What would transform the information environment is structural change in how political information is produced and distributed — public investment in trusted authentication infrastructure, changes to the algorithmic amplification systems that make synthetic content viral, and a political culture that treats the integrity of the information environment as a collective public good rather than a competitive advantage to be exploited.

Those are not legal reforms. They are political reforms, in the deepest sense. They require the people who benefit most from the current information environment — incumbents, well-funded campaigns, parties with sophisticated AI operations — to voluntarily constrain their own advantage.

History does not suggest that is likely. But the alternative is elections that are, in formal terms, free and fair, and in practical terms, shaped by whoever is most willing to automate deception.