The United States has no federal standard for AI deepfakes in political advertising. What it has instead is a checkerboard: twenty-nine states with deepfake election laws in effect, two states whose laws were struck down on constitutional grounds, and the rest with nothing at all.
According to Axios, the variety is not just in whether laws exist but in what they do. Minnesota and Texas prohibit political deepfakes for a set period before an election. Maryland bans them year-round. Colorado and Utah require detailed disclosures — who created the deepfake, when it was made, how it was edited. Other states simply require a label when AI is used in campaign ads.
And then there are California and Hawaii, where courts struck down deepfake laws on First Amendment grounds. Both states are trying alternative approaches that avoid directly regulating election-related AI content.
The First Amendment wall
California’s and Hawaii’s experience is the cautionary tale. Laws that specifically targeted political deepfakes were enjoined by courts that found them to be speech restrictions failing constitutional scrutiny. The lesson other states appear to have absorbed: regulate the disclosure, not the content.
Minnesota’s approach to nudification technology — the first state to ban tools that generate fake nude images of real people — faces a similar legal test. A judge denied xAI’s effort to halt the ban in July, but the broader constitutional question of whether regulating access to software (rather than penalising harmful conduct) survives First Amendment review remains unresolved.
The tension is structural. Content-based restrictions on AI-generated speech face a high bar under U.S. constitutional law. Disclosure requirements, which compel information rather than prohibiting expression, have an easier path — but even those have not been uniformly tested.
What Congress has done (and hasn’t)
The federal government’s first AI deepfake law addressed a different problem. The Take It Down Act, which took effect May 19, 2026, targets nonconsensual intimate imagery — a separate issue from election deepfakes, though one that has collided with political campaigns targeting women lawmakers.
Implementation has been uneven. Advocates cite inconsistent removal processes across platforms and predict litigation will be needed to secure meaningful enforcement. House Democrats have signalled plans to push election-specific deepfake legislation next year if they regain power.
Meanwhile, the patchwork deepens. States that acted early are refining their approaches. States that waited are studying what survived court challenges elsewhere. And voters in November will experience different realities depending on which state they cast their ballot in.
The disclosure vs prohibition split
The state laws fall broadly into two camps:
Prohibition states — Minnesota, Texas, Maryland — make it illegal to create or distribute certain political deepfakes. These face the steepest constitutional challenge because they restrict speech directly.
Disclosure states — the majority — require labels or disclaimers when AI is used in political advertising. These are easier to defend constitutionally but depend on compliance and enforcement. A disclosure requirement is only as good as the mechanism that catches violations.
The National Conference of State Legislatures tracks the full landscape. The picture changes monthly as new bills pass, get challenged, or are enjoined.
NZ angle
New Zealand’s deepfake legislation is at an earlier stage. The deepfake bill passed its first reading unanimously in 2026 but has not yet reached the committee of the whole House. The bill targets intimate-image abuse rather than election disinformation specifically.
The U.S. patchwork is a preview of the choices NZ lawmakers face. A disclosure-based approach — requiring AI-generated political content to be labelled — sidesteps the free speech arguments that have sunk blanket prohibitions in California and Hawaii. But disclosure regimes require enforcement infrastructure: a body to investigate violations, a standard for what counts as “AI-generated,” and penalties that mean something.
The other question is timing. U.S. states that passed deepfake laws close to elections faced legal challenges that took months to resolve — sometimes longer than the campaign itself. If NZ waits until an election cycle is underway to pass and implement rules, the courts may not have time to settle the questions before the votes are counted.
❓ FAQ
How many US states have AI deepfake election laws? Twenty-nine states have laws in effect as of August 2026, according to NCSL. Two states (California and Hawaii) had their laws struck down on First Amendment grounds.
What is the Take It Down Act? The first federal U.S. law addressing AI deepfakes, effective May 2026. It targets nonconsensual intimate imagery and requires platforms to establish removal processes. It does not address election deepfakes specifically.
What is the difference between prohibition and disclosure laws? Prohibition laws make it illegal to create or distribute certain deepfakes. Disclosure laws require labelling AI-generated content. Prohibition laws face higher constitutional hurdles; disclosure laws are easier to defend but depend on enforcement.
Does New Zealand have deepfake laws? A bill targeting intimate-image deepfakes passed its first reading unanimously in 2026. Election-specific deepfake regulation has not been legislated.
