Imagine sitting down to watch a campaign video in which a candidate looks straight into the camera and admits to taking a bribe. You recognize the face. You recognize the voice. It feels real, and it sounds real. But it is not real. It was manufactured by artificial intelligence, stitched together from bits of footage and audio to create a confession that never happened. Now imagine that your state has passed a law against AI-generated election deepfakes. Surely that means the video has to come down, right? Not necessarily. In June 2026, the National Conference of State Legislatures counted 31 states with election deepfake laws on the books. California and Texas were the first, enacting their laws back in 2019, but most states jumped on board in 2024 or later. Of those 31 states, 28 require some form of disclosure—usually a warning telling viewers that the content was generated or manipulated with AI. The other three, Maryland, Minnesota, and Texas, go further and prohibit certain deceptive election deepfakes outright, even if the content carries an AI warning. So the answer to whether that fabricated confession must be pulled from the airwaves depends entirely on where you live, what the law says, and whether the video fits the specific conditions lawmakers wrote into the statute. Policy researchers at the University of Denver, who track AI legislation across the country, have been watching these laws multiply. With the 2026 midterm elections approaching, the critical question for voters is not just whether a deepfake law exists, but what protection it actually provides in practice.
Louisiana offers a good example of the softer, disclosure-based approach. Under a law that took effect in June 2026, political campaign ads that use AI to falsely depict a candidate must carry a clear warning about that AI use. Imagine a campaign ad that puts a candidate’s face onto someone else’s body, making it look as though the candidate said or did something that never happened. If the ad is a message urging people to vote for or against a candidate, and it appears in printed materials, online advertisements, or broadcasts, then Louisiana requires a warning. And here is the key point: adding that warning can satisfy the state’s disclosure requirement. The fabricated ad can stay in circulation, as long as viewers are told that the content was made with AI. The warning tells voters how the content was created, not whether its accusations are true. It also does not excuse violations of other laws, like defamation or fraud. A 2023 Republican National Committee ad attacking former President Joe Biden illustrates this distinction. The ad depicted an imagined future after Biden’s reelection and carried a warning that it was “built entirely with AI imagery.” Although it predates Louisiana’s law, it shows how a disclosure can accompany an ad without stopping its circulation. But this protection does not cover every candidate on the ballot. The Louisiana law explicitly excludes candidates for federal office. So a fabricated video about a congressional candidate does not need an AI warning at all. For ads involving state and local candidates, leaving out a required warning can have serious consequences. Violators who are found to have damaged a candidate’s reputation or deceived voters can face a fine of up to two thousand dollars, up to two years in prison, or both. Local district attorneys generally decide whether to prosecute, subject to the state attorney general’s supervision. Louisiana also requires AI disclosures in certain campaign calls, including robocalls. Under a May 2026 law, calls that use AI to reproduce a public figure’s voice must disclose that use at the beginning of the call. The state’s board of ethics enforces the requirement, and violators can face civil fines of up to twenty-five hundred dollars for a first violation and five thousand dollars for subsequent violations. So in Louisiana, the warning is the shield: if you disclose the AI, the content can keep flowing.
Maryland takes a very different approach. Instead of allowing deceptive deepfakes to circulate with a warning, Maryland prohibits certain deepfakes entirely, even when they carry an AI label. The state’s May 2026 law covers images, audio, and video created or altered with AI or other digital tools to falsely depict a person in a way that looks or sounds genuine. Adding a label to that fabricated confession would not, by itself, make it lawful. Whether someone violates the law depends on their actions and purpose. The person must knowingly or recklessly create, use, or spread a deepfake to produce materially false information, with an intent such as influencing a voting decision. The law separately requires actual or intended harm to a voter, potential voter, or ballot petition, but it does not define what that harm means. Misleading voters about a candidate alone does not automatically establish a violation. A conviction can bring a fine of up to five thousand dollars, up to five years in prison, or both. Maryland also gives election officials new tools to respond when false voting information is spread. The state’s top election administrator must publicly correct credible reports of misinformation about voting procedures, results, or rights. For example, Maryland’s State Board of Elections has a rumor control page to monitor disinformation. The administrator can also seek court-ordered removal of misinformation, though not against online services that are simply hosting other people’s posts. These powers do not let election officials remove every false claim about a candidate. The public can report election misinformation, and officials can pass those reports to the state attorney general. Maryland’s reporting system dates back to a 2024 law, but its deepfake prohibition and new removal authority took effect on June 1, 2026. Neither guarantee guarantees that a correction will reach voters before they cast their ballots. The law gives officials a way to respond after the fact, but it does not create a perfect filter that keeps every deceptive deepfake out of the public square.
How much does a warning actually help? That is a crucial question, because most states rely on warnings rather than outright bans. By early September 2026, the Wesleyan Media Project had identified at least 164 political ads nationwide that were created or enhanced with AI during the 2026 election cycle, accounting for nearly eighty million dollars in ad spending. About seven in ten of those ads carried no AI disclosure. The count includes uses beyond deceptive impersonations, and it does not establish how many ads violated a law or actually misled voters. But it gives a sense of how quickly AI has entered political advertising. Research on warning labels offers some hope, but also some caution. A 2025 study tested warning labels on misleading AI-generated images in two experiments involving 7,579 Americans. People who saw the labels were less likely to believe the posts’ claims. That sounds encouraging. However, warnings that simply identified content as AI-generated did little to change how willing people said they were to engage with it, including sharing it. In other words, a label might make you doubt a post, but it might not stop you from passing it along. Warning labels can also raise doubts about accurate information. Another recent study found that people rated headlines as less accurate when they were labeled as AI-generated, regardless of whether the headlines were true or false. Knowing that AI helped produce something does not tell a reader whether its claims are correct. A true story generated with AI assistance is still true, but the label can make it seem suspicious. Neither study tested whether a state’s disclosure law changes how people vote. The studies also cannot tell us whether the threat of punishment under Maryland’s law discourages people from creating or spreading deepfakes. So there is a real gap between what these laws intend to do and what they actually accomplish in the messy, fast-moving world of online politics.
Passing a law also does not settle whether officials can enforce it. In September 2026, a federal judge temporarily barred Montana from enforcing its deepfake law, finding that it likely posed a threat to a conservative political action committee’s free speech rights under the First Amendment. That case is a reminder that deepfake laws must be written carefully to avoid running into constitutional protections for political speech. Even when laws survive legal challenges, there are gaps in what they cover. For voters, the most important gap is this: a video without an AI warning has not necessarily passed an authenticity test. The disclosure rule might not cover it at all, as with congressional candidates under Louisiana’s provision. Or the video might be covered, but the person who created it simply ignored the law. A warning’s absence cannot tell viewers which explanation applies. You cannot look at an unlabeled video and know whether it is genuinely authentic, whether it is AI-generated but legally exempt, or whether it is AI-generated and illegal but still circulating because no one has stopped it. The same is true in the other direction. A video with an AI warning is not necessarily false. It might be a legitimate ad that used AI for editing or enhancement, or it might be a deceptive deepfake that is legal in a disclosure state but illegal in a prohibition state. The warning tells you something about how the content was made, but it does not tell you whether the underlying claims are true. This creates a confusing environment for voters who are trying to make sense of what they see on their screens.
So let’s go back to that video of a candidate apparently admitting to taking a bribe. Under both Louisiana’s and Maryland’s laws, it could still reach voters. In Louisiana, if the video is covered by the disclosure rule, it can circulate freely as long as it carries the required warning. In Maryland, creating or spreading it could lead to punishment, but only if the law’s conditions are met: the person acted knowingly or recklessly, the content was materially false, there was an intent to influence a vote, and there was actual or intended harm. If those conditions are not met, the video might not violate the law at all. And even if it does violate the law, enforcement takes time. A court order might come after the election is over. Neither approach guarantees that voters will avoid seeing the fabricated confession before they cast their ballots. These laws give states meaningful ways to respond to deception, and they signal that lawmakers are paying attention to the threat of AI-driven misinformation. But their passage alone cannot guarantee voters that what they are watching is real. In the end, the most honest takeaway is a humble one: in the age of AI, seeing is no longer believing. A warning can help, but it is not a cure. A missing warning is not proof of authenticity. And a law, no matter how well-intentioned, cannot fully protect you from the power of a convincing lie. The burden of skepticism, unfortunately, falls on each of us.