States Are Building a Patchwork of Laws to Regulate Election Deepfakes

Last updated 5 days ago. Our resources are updated regularly but please keep in mind that links, programs, policies, and contact information do change.

Artificial intelligence (AI) can make a candidate appear to say something the candidate never said. States are answering these election deepfakes with rules that can require a warning label, let a candidate sue or expose the sender to criminal penalties. The hard part is drawing a boundary between deceiving voters and making a political joke, while giving senders a rule they can understand.

A convincing fake is only the start of the legal question

The National Conference of State Legislatures counted 31 states with enacted laws regulating deepfakes in political messaging in its inventory updated June 23, 2026. That number describes legislation, not a nationwide ban or a guarantee that every provision can be enforced.

The inventory divides those laws into prohibitions and disclosure requirements, including year-round restrictions in Maryland and election-related restrictions in Minnesota and Texas. The useful question is which rule applies to a particular message, not whether a state appears on the list.

Washington’s definition requires intentionally manipulated digital media that realistically depicts something a real person did not do or say and creates a fundamentally different impression from the original. Minnesota’s definition likewise asks whether a reasonable person would believe the depicted speech or conduct actually happened, and distinguishes technical production from physical or verbal imitation. The legal target is a false appearance of authenticity; the presence of editing technology alone does not answer the question.

The same message can face different state rules

A label requirement, a lawsuit and a criminal offense impose different burdens on the sender.

These are selected examples, and California’s text must also be read alongside the court history below. The comparison explains why a checklist limited to “AI-generated” and “near an election” would miss important conditions.

California’s section 20012 defines malice as knowing the deceptive content was false or acting with reckless disregard for the truth. Here, malice concerns the sender’s relationship to the truth, rather than ordinary personal hostility. Materiality asks whether the deception matters to voting decisions or confidence in the election process or result.

California approved Assembly Bill 502 on September 19, 2026 as an urgency statute amending section 20012 with immediate effect. Older descriptions of that provision’s election window and parody exception therefore need particular care.

What a disclosure must actually do

Washington provides a defense when the message carries the disclosure “This (image/video/audio) has been manipulated,” presented in the format the statute requires. A defense gives a sender an answer to a claim under that statute; it does not eliminate every possible legal claim about the message.

For visual media, Washington requires easily readable text no smaller than the largest other text, and a video must show the disclosure for its entire duration. For audio alone, the disclosure must be clearly spoken at the beginning and end, and at intervals no longer than two minutes if the recording exceeds two minutes. A small caption outside the video or a notice heard only after a long recording can miss the specific format requirements.

Minnesota’s statute provides no prescribed AI-label safe harbor, although its prohibition excludes federally required broadcasts of candidate-produced deepfakes. That does not mean a label is irrelevant: a label can help establish that a viewer would understand a parody rather than believe a false event occurred.

California’s amended section 20012 exempts content a reasonable person would understand as satire or parody, or content with the specified manipulation-for-satire-or-parody disclosure.

Washington excludes qualifying news, features, commentary and editorials in regularly scheduled news media not controlled by a candidate or a political or incidental committee. California’s exception for qualifying news broadcasts and commentary requires clearly acknowledging that the deceptive material does not accurately represent a real event or speech.

Washington generally places liability on the sponsor, but a disseminating medium can become liable for removing a disclosure or changing the content into synthetic media, subject to specified defenses and the federally required candidate-broadcast exception. Reposting can therefore change the legal analysis if the repost strips away the information that made the original message identifiable as manipulated.

Enforcement can mean removal, damages or punishment

In Washington, the depicted candidate can seek a court order prohibiting publication or sue the sponsor for general or special damages, and the court may award a prevailing party attorney’s fees and costs. Those tools address different needs: stopping circulation and seeking compensation after harm.

The Washington plaintiff must establish use of synthetic media by clear and convincing evidence, and the statute encourages courts to act expediently. An expedited expectation is not a promise that a disputed fake will disappear before everyone sees it.

Washington appellate authority describes clear and convincing evidence as showing the disputed fact is highly probable. The standard requires a strong showing, rather than simply pointing to something suspicious in a clip.

The Washington Public Disclosure Commission explained in May 2024 that it would not investigate complaints under this synthetic-media law, which creates a private civil cause of action. Identifying the correct enforcement route matters as much as identifying the right statute.

Minnesota allows its attorney general, county or city attorneys, the depicted individual, and an injured or likely injured candidate to seek injunctions against a violation underway or reasonably believed imminent. A prospective court order can address a threatened distribution before the separate criminal process reaches a verdict.

Minnesota’s ordinary penalty is up to 90 days in jail, a $1,000 fine or both; an offense within five years of a prior conviction under this statute can bring up to five years and $10,000, while intent to cause violence or bodily harm carries up to 364 days and $3,000. Texas classifies its election deepfake-video offense as a Class A misdemeanor, a criminal offense.

A convicted Minnesota state or local candidate forfeits the nomination or office, and the statute also disqualifies convicted individuals from its specified appointments. These penalties are potential consequences of proving an offense, not proof that any particular manipulated message meets the elements.

A robocall shows why the channel matters

Before New Hampshire’s 2024 presidential primary, thousands of homes received a robocall using a voice resembling President Joe Biden’s and urging recipients to skip the primary and save their vote for November, New Hampshire Public Radio reported. The deception involved not only the apparent speaker but a direct instruction about participating in an election.

The Federal Communications Commission (FCC) determined that organizer Steve Kramer knowingly caused calls with misleading caller identification and intent to defraud and cause harm, violating the Truth in Caller ID Act. The FCC imposed a $6 million forfeiture on September 30, 2024 for that caller-ID violation. A forfeiture is an assessed administrative penalty; its announcement does not establish that the money was collected.

New Hampshire Public Radio reported on June 13, 2025 that a Belknap County Superior Court jury had acquitted Kramer on all charges in the separate criminal case. The FCC finding and the criminal acquittal concern different proceedings, so neither should be substituted for the other.

Protecting elections also raises speech questions

When signing California’s earlier measures in September 2024, Governor Gavin Newsom argued that election integrity and public trust required protection against AI disinformation. In AB 502’s findings, the legislature argues that voters focus on elections in the weeks before an election and that disinformation may not be corrected before they cast their votes. That rationale treats convincing impersonation as a threat to informed participation, rather than merely an unpleasant political argument.

The California legislature argues that limited election periods and labels identifying inauthentic media narrowly target voter deception. The constitutional question is whether the actual restrictions stay within that stated purpose without unnecessarily burdening political expression.

The Hamilton Lincoln Law Institute, representing video creator Christopher Kohls, describes his July 26, 2024 Harris-voice video as a labeled parody and argues that the earlier California law’s oversized disclaimer requirement and vague standards chilled political satire. For that challenge, identifying the content as a joke was part of the speaker’s defense, not an admission that it should be treated as authentic campaign information.

Kohls v. Bonta is the institute’s constitutional challenge to California’s AB 2839 and its separate platform-focused AB 2655 in the U.S. District Court for the Eastern District of California.

The federal district court reviewing the earlier AB 2839 concluded that its required disclaimer could occupy an entire screen and that forcing such a disclaimer on parody could destroy the joke. An accurate warning can still impose a speech burden when its size or placement overwhelms the expression it accompanies.

On August 29, 2025, that district court permanently enjoined California officials from enforcing AB 2839 against the named plaintiffs. The order’s party-specific remedy and the later September 2026 amendment make “California’s deepfake law is banned” an inadequate description of the legal position.

The 2025 order addressed the earlier text; AB 502’s enactment does not itself determine how that injunction applies to a message governed by the amended text. That interaction requires checking the particular court remedy, rather than assuming either that the amendment removed the injunction or that the judgment settled the revised law’s constitutionality.

Kohls v. Ellison is the federal constitutional challenge to Minnesota’s election-deepfake statute brought by the Hamilton Lincoln Law Institute and Upper Midwest Law Center. On February 9, 2026, the U.S. Court of Appeals for the Eighth Circuit upheld the refusal to temporarily block enforcement. The Eighth Circuit left possible later standing, merits and permanent-relief questions for the district court. That procedural result should not be reported as a final decision that every application of Minnesota’s law is constitutional.

The institute’s later case-page update reports that rehearing was denied and the case was being held for possible coordination with X Corp.’s similar challenge. The update does not give a date for those later steps, so it should not be read as a new dated merits ruling.

Federal rules reach particular conduct

In a ruling released February 8, 2024, the FCC confirmed that AI-generated human voices fall within existing Telephone Consumer Protection Act restrictions on artificial or prerecorded voices, requiring prior express consent absent an emergency purpose or exemption. The ruling applies an existing telephone framework to a new way of making a voice; it does not create a rule for every fake campaign image or online video.

The federal rules include a noncommercial residential-call exemption for no more than three calls in a consecutive 30-day period when the required opt-outs are honored. Consent requirements and exceptions have to be checked for the kind of number called, rather than assuming all political AI calls are treated alike.

The FCC also proposes tighter telephone-provider filings and screening in its Robocall Mitigation Database, part of its framework for preventing illegal calls. Under the comment schedule set in the FCC’s September 9, 2026 notice, comments are due October 9, 2026, and reply comments are due November 9, 2026.

Federal election law prohibits a candidate or agent from falsely purporting to act for another candidate or political party in a damaging way, and separately prohibits anyone from fraudulent impersonation to solicit contributions. On September 19, 2024, the Federal Election Commission declined to open an AI-campaign-ad rulemaking and instead adopted an interpretation explaining that these fraud rules apply regardless of technology. The federal layer therefore targets specified conduct, such as deceptive caller identification or campaign impersonation, rather than supplying one answer to every misleading election message.

Labels are measurable, but effectiveness is a different question

The Wesleyan Media Project reported on September 4, 2026 that it had identified at least 164 AI-generated or AI-enhanced political ads. Of the ads it identified, 69 percent lacked a disclaimer disclosing AI use. That observation shows labels were far from universal in the identified advertising, rather than establishing that every unlabeled ad violated a law.

A prevalence count does not measure whether a particular statute reduced voter deception or changed an election outcome. The existence of a penalty likewise shows that an enforcement tool was used, not that it prevented the original message from reaching voters. Evaluating the laws requires distinguishing disclosure, legal coverage and actual effects on voters.

Start with the rule and the official voting information

For a questionable message, separate the factual verification from the legal test. The National Association of Secretaries of State’s January 2026 TrustedInfo initiative directs the public to state and local election officials for reliable election information. Check voting instructions with those officials, especially if an apparently familiar voice announces a changed location or tells voters to stay home.

Report an unwanted robocall through the FCC Consumer Complaint Center. The FCC’s imposter-call guide directs users to the “Phone” form and “Unwanted Calls/Texts” issue, and asks for the displayed caller-ID number, the number called, the call time and time zone, and any recording or screenshot. Preserve those details when reporting a suspected fake call; an FCC complaint and a state-law court action follow different enforcement routes.

For the legal question, check the current statute and any court order affecting the provision. Identify the depicted person, the medium, the election period, the required intent and any applicable disclosure or exception before choosing an enforcement route. That is how the patchwork becomes usable: a specific message, matched to a specific rule, rather than a national count treated as a verdict.