Thirty-one states, three courtrooms
Voters in 31 states cast ballots on Tuesday under a state law on AI deepfakes in election communications, by our count. Courts have struck down Hawaii's, blocked one of California's and left Minnesota's standing for now.
Asterzenback, CC0, via Wikimedia CommonsThirty-one states have a law on the books regulating AI-generated deepfakes in election communications as voters cast ballots in the US midterm elections on Tuesday, by Komposite's count of Public Citizen's legislative tracker, last updated on 21 October. Thirty-three states have passed such a law at some point. Massachusetts let its law lapse in February 2025, and a federal judge struck down Hawaii's in January.
A deepfake, in these statutes, is audio, video or an image altered or generated with digital tools so that it shows a real person saying or doing something they did not. The laws differ in scope and penalty. Minnesota's makes it a crime to spread one within a set period before an election, without the depicted person's consent and with intent to injure a candidate or influence the result. Most of the laws are recent. By our count, only five states had one before 2024; sixteen added a law that year, seven in 2025 and five so far in 2026, the most recent Louisiana in June. When Maryland's law was signed in May, Public Citizen counted it as the 30th, and said that "the need for the remaining 20 states to enact protections has never been greater". By our arithmetic, 19 states now have none.
| First law enacted | States | Count |
|---|---|---|
| 2019 to 2023 | California, Texas (2019); Michigan, Minnesota, Washington (2023) | 5 |
| 2024 | Alabama, Arizona, Colorado, Delaware, Florida, Hawaii, Idaho, Indiana, Massachusetts, Mississippi, New Hampshire, New Mexico, New York, Oregon, Utah, Wisconsin | 16 |
| 2025 | Kentucky, Montana, Nevada, North Dakota, Pennsylvania, Rhode Island, South Dakota | 7 |
| 2026 | Tennessee, Maine, Vermont, Maryland, Louisiana | 5 |
| Enacted at some point | 33 | |
| Less: lapsed or struck down | Massachusetts (sunset February 2025); Hawaii (struck down January 2026) | −2 |
| On the books on 3 November | 31 |
The laws have fared poorly when tested in court. Hawaii's Act 191 barred the reckless distribution of "materially deceptive media" that risked harming a candidate or changing how people voted, from the first working day of February in every even-numbered year through the general election. A disclaimer shown throughout a video or read at the start and end of audio was a safe harbour. On 30 January Judge Shanlyn Park of the US District Court for Hawaii held in a case brought by the satirical site the Babylon Bee that the law failed strict scrutiny and was unconstitutionally vague, The Volokh Conspiracy reported. Counter-speech, voter education and existing defamation and election-fraud laws were less restrictive options, she wrote, adding: "The First Amendment does not permit the State to sacrifice speech for efficiency."
California's 2024 law fared no better. A federal judge granted a preliminary injunction against AB 2839 in October 2024, two weeks after it was signed, and on 29 August 2025 granted the challengers summary judgment, holding the statute unconstitutional in its entirety, according to the case record kept by the Hamilton Lincoln Law Institute, which represents one of the challengers. The court did not rule on a companion law, AB 2655, because California had agreed not to enforce it. The state appealed to the Ninth Circuit on 19 September 2025; the briefs were filed in January and March, and no decision has been reported. By our arithmetic, the appeal has been pending for 410 days.
| State and law | Case | What the court did | Where it stands |
|---|---|---|---|
| Hawaii, Act 191 (SB 2687, 2024) | Babylon Bee v. Lopez, US District Court for Hawaii | Judge Shanlyn Park held on 30 January 2026 that the law fails strict scrutiny and is unconstitutionally vague, and rejected its disclaimer safe harbour | Struck down |
| California, AB 2839 (2024) | Kohls v. Bonta, US District Court for the Eastern District of California | Preliminary injunction on 2 October 2024; summary judgment for the challengers on 29 August 2025, finding the statute unconstitutional in its entirety | Blocked; California's appeal to the Ninth Circuit pending since 19 September 2025 |
| California, AB 2655 (2024) | Kohls v. Bonta | Not ruled on: California had agreed not to enforce it | Not enforced; part of the same appeal |
| Minnesota, 2023 law as amended in 2024 | Kohls v. Ellison, Eighth Circuit | On 9 February 2026 upheld the refusal of a preliminary injunction, on standing and delay, without deciding the constitutional question | In force; the challenge continues in the district court |
Minnesota's law is the exception so far. On 9 February the Eighth Circuit upheld a lower court's refusal to block it, The Volokh Conspiracy reported. The court found that one challenger, whose videos were labelled as parody, lacked standing because labelled parodies are not deepfakes under the statute, and that the other had waited 16 months to sue. It expressly did not decide whether the law is constitutional, and the case continues. Its penalties include one aimed at candidates: one who breaks the law forfeits the nomination or office.
What this means for the technology industry is narrower than the headline count suggests. The laws we reviewed are aimed at whoever distributes a deepfake, not at the AI companies whose tools can produce one. The court record so far points in one direction: of the three states whose laws we found tested on the merits or close to it, two have lost, and the third has won only on procedure. A disclaimer, which many states treat as the safe option, did not save Hawaii's law or California's.
Why it matters: election deepfake rules are set state by state, so a clip that is lawful to share in one state can break the law in its neighbour. Deepfakes are already a fraud problem for companies, which have rebuilt payment verification around cloned voices, and state officials have been willing to act on AI on their own, as when eight attorneys general sued OpenAI last month over ChatGPT's safety. On the money behind the election, Komposite reports separately on the two networks of AI-funded super PACs.