In the quiet machinery of democracy, misinformation often looks less like a science-fiction nightmare and more like a misleading date on a campaign flyer. That is the reality Maryland election officials have confronted since a new state law targeting AI-generated election disinformation took effect in June. In the months since, the State Board of Elections has received five formal complaints under the law, and only one has been confirmed as a violation: a campaign advertisement that gave voters the wrong date for Election Day. No deepfake video of a candidate collapsing, no fabricated audio of a politician saying something racist, no viral image of a ballot box being tossed into a dumpster—just a simple, dangerous error about when to show up and vote. It is a reminder that the most insidious election lies are often the simplest ones, and that the technology used to spread them has made it easier than ever to manufacture confusion. The new law gives the board limited authority to combat false information about the “time, place, or manner” of elections, as well as results and voting rights. But it does not cover every kind of disinformation, and its penalties—up to five years in prison or a $5,000 fine—may not be enough to deter campaigns desperate for an edge. For voters, the stakes are straightforward: false information about when, where and how to vote can create real chaos, causing people to miss deadlines, show up at the wrong polling station, or lose faith in the process altogether. Maryland is not alone in wrestling with these issues, but its experience so far offers an early, imperfect glimpse of how American democracy is trying to adapt to a world where seeing is no longer believing. The complaints themselves are a sign that people are paying attention—that a candidate’s post, a Facebook image, a suspicious advertisement can now be flagged and reviewed. But they also reveal how much of this new battlefield remains unpoliced, and how easily a fictional image can slip past the gatekeepers and into the minds of voters. For every complaint filed, there may be many more falsehoods that never get noticed, either because they are too subtle or because voters simply do not know where to report them.
The law was written with an eye to the First Amendment, which makes any attempt to regulate political speech a delicate legal dance. It does not ban all deepfakes or all AI-generated content; it focuses narrowly on false information about the mechanics of voting. State Board of Elections spokesperson Lexy Rhoden said the board reviews each complaint, consults with the attorney general’s office, and tries to secure the removal of inaccurate content or issue public corrections. But if a complaint falls outside the law’s limited scope—say, a fabricated video of a candidate accepting a bribe—the board forwards it to the state prosecutor’s office. That office declined to comment on any ongoing investigations, citing standard policy. For legal scholars, the challenge is figuring out where to draw the line between protected political speech and actionable deception. “You can limit a whole host of things that are not within the context of the First Amendment,” said Donald Tobin, an election law professor at the University of Maryland. “But in this case, what you’re really trying to separate from is what I’ll call ‘artistic creativity’ and the First Amendment.” Tobin noted that the law requires a deepfake to be made with intent to harm someone before it can be penalized, a requirement he expects to be tested in court. Lawmakers, he said, were trying to balance the urgent need to regulate harmful uses of AI with the constitutional protection of even offensive or dishonest political speech. That balance is inherently fragile. A provision that is too broad risks chilling legitimate satire, parody, and even hard-hitting campaign attacks. A provision that is too narrow leaves dangerous falsehoods free to circulate. The result is a law that is at once historic and hesitant, a recognition that the same technology which can create a convincing fake video of a candidate can also be used to make an opponent look like a monster. For now, Maryland’s board can only act when the lie is tied to the nuts-and-bolts of voting—the dates, the hours, the locations, the rules. Everything else falls into a legal gray zone, where election officials can do little more than watch, forward, and hope someone else takes action.
A few weeks after the law took effect, the limits of that gray zone came into view. Ed Hale, a former Republican gubernatorial candidate, posted an AI-generated photo depicting his then-primary rival, Dan Cox, embracing Democratic Governor Wes Moore. The image was obviously manufactured—a political fever dream rather than a photograph—but it fed a narrative that could influence primary voters. Moore’s spokesperson said the post was reported to the State Board of Elections, though the board has not offered a public update. Then, earlier this month, Moore’s lawyers filed another complaint, alleging that an advertisement misused the governor’s likeness to make voters think he opposed the November redistricting ballot measure. Again, the board remained quiet about its status. These incidents show how easily AI can be weaponized in political campaigns, but they also show how hard it is for a narrow law to catch up. In Anne Arundel County, the issue hit even closer to home for voters accustomed to reading the Capital Gazette. A Facebook post from Chick and Ruth’s Delly, a beloved Annapolis restaurant owned by District 7 Democratic candidate Spencer Jones, featured an AI-altered version of the newspaper’s front page. The doctored image described Jones as the incumbent—he is a newcomer—and added a glowing line about the restaurant being a “quintessential Annapolis business” that draws people from around the region and country. The real story did not say that. Jones said an ad agency managing the restaurant’s social media had used AI to improve the clarity of a photo of the actual front page, and the software “hallucinated” the additions. He apologized, removed the post, and issued a corrected version. The episode is a textbook case of how AI can blur the line between enhancement and fabrication. No one set out to lie; the technology simply made it easy. But in an election, good intentions do not matter. What matters is what voters see, believe, and act on. The fact that Jones corrected the mistake quickly does not erase the fact that the false image circulated, and it underscores how easily even a local campaign can stumble into the brave new world of algorithmic deception.
The deeper question is whether the law’s penalties actually discourage campaigns from using AI to mislead. Candace Turitto, director of the University of Maryland’s Applied Political Analytics Program, is skeptical. A fine of $5,000 is pocket change for a well-funded campaign, and even a multiyear prison sentence is unlikely in a complaint that might not be resolved until after the election has already been decided. The potential benefits, by contrast, can be enormous: a misleading ad that goes viral might attract national media coverage, shape voter perceptions, and perhaps tip a close race. Turitto described a hypothetical candidate calculating that the upside of a deceptive deepfake outweighs the downside of a possible fine or legal action. “For the right kind of candidate in the right setting, maybe that’s a risk worth taking,” she said. That risk calculation is especially dangerous because the most damaging misinformation often spreads fastest in the final days before an election, when there is no time for a correction to catch up. A story can be debunked, but the debunking rarely travels as far or as fast as the original lie. And if the punishment arrives months or years later, it does nothing to help the voter who stayed home on the wrong day or who believed a fabricated scandal that changed their vote. The law also depends heavily on the willingness of campaigns and ordinary citizens to report suspicious content. That places a burden on voters who may not realize they are being misled, and on election officials who are already stretched thin. Turitto’s point is not that the law is useless; it is that deterrence is a game of incentives, and right now the incentives may be skewed toward the bad actors. Until the consequences are swift and certain—until a campaign knows it cannot profit from deception, even after the votes are counted—the threat of AI-generated disinformation will remain a shadow over every election. The five complaints in Maryland are a tiny sample, but they are enough to reveal the problem: the law exists, the technology exists, and the will to exploit it almost certainly exists too.
While state officials wrestle with policy, local election directors are living the day-to-day reality. Of the nine local election directors contacted by the Baltimore Sun, only four responded, and none said they had reported an instance of AI-generated election disinformation this season. That could mean the problem has not reached their communities, or it could mean it is hiding in plain sight. Gary Holmes, Cecil County’s election director, said he has seen AI-generated social media content promoting campaigns but nothing he considered disinformation worthy of a state complaint. Ruie Marie Lavoie, Baltimore County’s election director, said she was surprised her office had not encountered AI-generated disinformation, given her fears that the technology could spread false claims more quickly than ever. Her office has, however, dealt with old-fashioned misinformation: a false claim that voters had to pay postage to return mail-in ballots. The rumor prompted residents to call with questions, and officials had to correct it before it caused anyone to miss the deadline. There is an irony in that: the lie was not created by a machine, but it spread in exactly the way an AI-generated lie might, moving from social media post to anxious phone call to local news. Lavoie said she continues to monitor social media and encourages voters to verify anything they see with their local election office. Erin Perrone, Carroll County’s election director, urged voters to rely on local election offices for information about dates, polling places, and hours. “That’s our job,” she said. “That’s our job to help voters.” Allison Murphy, Caroline County’s election director, noted that voters who encounter suspected misinformation can report it through the State Board of Elections website, which has a form for false information about elections, results, and voting rights—regardless of whether AI was involved. These local officials are not technologists or legal scholars; they are public servants who answer phones, update websites, and try to make voting as smooth as possible. They are also the first line of defense against a problem that is bigger than any one county. Their experiences suggest that the AI disinformation crisis may not yet be visible in every jurisdiction, but the infrastructure for dealing with it—the watching, the reporting, the correcting—is already being built, one phone call at a time.
Maryland’s first encounter with AI election disinformation is less a story of dramatic deepfake scandals than a quiet, worrying reminder of how fragile democratic trust really is. The single violation confirmed so far—the campaign ad that gave voters the wrong Election Day date—may seem minor next to a fabricated video of a candidate admitting to a crime. But it is the kind of lie that has a direct, practical effect on whether someone can cast a ballot. And it is the kind of lie that, in the past, might have required a human to invent and distribute it. Now, artificial intelligence can produce a convincing falsehood at scale, tailored to specific communities, in a matter of seconds. The law’s narrow scope is a testament to the difficulty of writing rules for a technology that evolves faster than the legislative process. The questions it raises—what intent is required, what counts as artistic creativity, how to balance free speech and order, how to punish a campaign after an election has passed—are not unique to Maryland. They will be answered, slowly, through court challenges and legislative tweaks, through the work of local election officials and the vigilance of voters. But the answers cannot come soon enough. Every misleading date, every fabricated headline, every AI-generated embrace between political enemies chips away at the assumption that elections are fair and understandable. Restoring that assumption requires more than laws; it requires a shared commitment to truth, a willingness to verify, and a recognition that democracy depends on citizens being able to trust what they see and hear. The five complaints in Maryland are a small number, but they are a signal. They show that officials are watching, that the public is reporting, and that the legal system is struggling to catch up. In the end, the fight against AI disinformation is not just about penalties or policies. It is about the simple, stubborn faith that an election is a place where facts still matter—and where a vote cast on the wrong day should never be the result of a machine’s lie. Maryland has taken the first step. The journey is just beginning.