The woman on the screen looks like someone you might know—or wish you did. She is Dona Maria, a straight-talking, middle-aged Black woman with a sharp tongue and a no-nonsense attitude. In her videos, she looks directly into the camera, her voice rising with frustration as she rails against the policies of Brazilian President Luiz Inácio Lula da Silva. “They line their pockets with our sweat!” she declares in one widely shared clip about rising electricity bills. Her anger feels visceral, her words resonate with the kind of everyday bitterness that spreads quickly in a country where inequality is a daily reality. Millions of Brazilians watched her, shared her videos, and nodded along. But Dona Maria does not exist. She is a hyperrealistic avatar, a creation of artificial intelligence, designed to look and sound so convincingly human that she became a political phenomenon overnight. Her rise is the perfect symbol of Brazil’s newest electoral dilemma: a nation bracing for one of the most consequential votes in its recent history, while fighting an invisible enemy that can fabricate reality itself.
Brazil is not alone in this struggle. Around the world, from India to Germany to Hungary, democracies are wrestling with how to keep AI from poisoning the well of public discourse. But Brazil feels the pressure most acutely. It is a hyper-online country, a place where WhatsApp messages can topple governments and where conspiracy theories spread faster than wildfire. In 2018, the nation was “bulldozed by disinformation,” as Camila Tsuzuki, head of research at the nonprofit InternetLab, puts it. Back then, fake news and fabricated images portraying leftist candidates as pedophiles or agents of “gender ideology” flooded private messaging apps, helping to elect the far-right populist Jair Bolsonaro. Four years later, when Bolsonaro narrowly lost his reelection bid to Lula, his supporters refused to accept the outcome, echoing baseless claims of electoral fraud. That rage culminated in the Jan. 8, 2023, attacks on the capital’s government buildings, a violent insurrection that shocked the world and left deep scars on Brazilian democracy. Bolsonaro himself is now serving a 27-year sentence under house arrest for his role in plotting that coup attempt. It is no wonder that Brazil’s electoral authorities have taken a proactive, even aggressive stance against AI abuse. The Superior Electoral Court passed sweeping rules requiring AI-generated campaign material to be clearly labeled, banning the use of AI to impersonate candidates, and prohibiting the distribution of synthetic content that could mislead voters. These are among the strictest regulations anywhere, and they were enacted with a sense of urgency that many other nations have lacked.
On the streets of Rio de Janeiro, the fear of artificial deception is palpable. People like Ana Luiza Lelis, a campaign worker who has spent her days passing out leaflets for a gubernatorial candidate, understand the danger intimately. “If you don’t pay attention to the detail, you’ll believe what you see,” she said, pausing on her way home through the city’s crowded streets. Her words capture the anxiety of a population that has learned to distrust what it sees and hears. The rise of AI-generated avatars like Dona Maria deepens that unease because they exploit a fundamental human weakness: our tendency to trust faces and voices that feel familiar. These “synthfakes,” as some researchers call them, are not just cheap tricks; they are sophisticated psychological tools. Matheus Soares, the head of content at Aláfia Lab, a civil society organization that co-runs a monitoring project tracking AI in the elections, explains that not all AI use is harmful, but cases like Dona Maria are especially troubling. “These are very realistic characters commenting on politics,” he says. “They can have an impact on public debate, on citizens’ perceptions of a candidate or a political topic.” The physical appearance of these avatars matters enormously. As Tsuzuki notes, “Physical characteristics create a rapport with the person receiving the information. And the greater the rapport, the greater the likelihood that someone will be open to the information or the persuasion of that message.” Dona Maria’s creators clearly understood this. Her face was designed to evoke the countless Black women who form the backbone of Brazilian communities, women whose wisdom and distrust of authority are often treated as common sense. By channeling that archetype, Dona Maria became a trusted voice, even though her true identity and motives remain unknown. Lula’s coalition filed a legal complaint demanding her takedown, arguing that the anonymous accounts behind her peddled false and distorted information to mislead voters. Her main accounts on TikTok and Instagram were eventually removed in May, but like a hydra, new secondary accounts have popped up, and fragments of her content still echo across the internet. Regulators have discovered a bitter truth: in the digital age, banishing a lie is like trying to erase a rumor from memory—impossible.
Candidates themselves are navigating this treacherous landscape with varying degrees of caution. Rafael Satiê, a Rio de Janeiro city councilor running for federal deputy on a pro-Bolsonaro platform, prefers to steer clear of generative AI altogether. “I prefer not to use it, to avoid running into any issues,” he said, flanked by a crew of real-life social media influencers as he headed into a campaign event. His wariness is shared by many politicians who fear that even innocent AI use might be misconstrued or punished by the electoral court. Brazil’s strict regulatory environment owes much to the unique power of its Superior Electoral Court, which can unilaterally pass resolutions and rule on disputes without waiting for a slow-moving legislature. That institutional agility has made Brazil a global pioneer, but it also highlights how far behind other countries remain. In the United States, for example, at least 31 states have enacted local legislation targeting deceptive AI in elections, but bipartisan federal bills have repeatedly stalled in Congress. J.B. Branch, director of federal AI governance and technology policy at the U.S. consumer advocacy organization Public Citizen, notes that globally, election-specific AI regulation has “been perceived as a back burner issue” compared with other pressing threats like autonomous weapons or job displacement. Meanwhile, Brazil’s broader AI regulation bill remains stuck in the lower house, leaving the electoral court as the only line of defense. “Trying to regulate individual users is borderline impossible,” Branch admits. The endless, iterative nature of online content makes enforcement a game of whack-a-mole. Take down one account, and three more appear. The monitor project, run by Aláfia Lab and InternetLab, has recorded 29 cases of AI avatars like Dona Maria since the campaign began, though none achieved her level of notoriety. An analysis by BBC News Brasil found that in April, Dona Maria’s content was generating engagement comparable to that of national politicians—a stunning testament to the power of a well-crafted illusion.
The next frontier is even more insidious. Beyond synthetic avatars, researchers are now focusing on the influence of chatbots—systems like ChatGPT, Anthropic’s Claude, and Google’s Gemini. These tools can hold conversations, answer questions, and, according to a growing body of scientific research, shift people’s political opinions. Experiments in the United Kingdom and the United States have shown that chatbots can subtly nudge users toward certain viewpoints, often without the users realizing it. Brazil’s electoral laws attempt to address this by banning AI tools from ranking candidates or emitting opinions about them. But a study conducted by InternetLab in late August found that several chatbots operating in Brazil are already doing exactly that, ignoring the legal prohibitions with impunity. This raises a deeply uncomfortable question for the 159 million Brazilian voters who will head to the polls on Oct. 4: If their final decision is being shaped by invisible algorithms, are they still making a choice of their own? To mitigate the chaos, Brazil’s electoral court has imposed a blackout period—an absolute ban on posting or resharing AI-generated political content for 72 hours before the vote and 24 hours after. The idea is to give voters a moment of calm in which they can reflect without being bombarded by synthetic influence. Yet chatbots remain outside that blackout, quietly continuing their conversations with voters right up until the polls close. Tsuzuki calls this “the big question of these elections,” predicting that the impact of these processes and platforms on voters’ decision-making will prove to be a stress test for democracy itself.
As Brazilians prepare to choose their next president, senators, governors, and lawmakers, the nation stands at the edge of an unprecedented experiment. It is an election in which the tools of persuasion can be manufactured out of thin air, in which the face of a trusted neighbor might be a digital phantom, in which every voice—human or machine—demands equal attention. Lula and Senator Flávio Bolsonaro, a son of the former president, are locked in a neck-and-neck race, and the outcome could hinge on how well ordinary citizens can separate truth from fabrication. The human spirit, however, is not so easily defeated. In the crowded streets of Rio, in the humble homes where families watch candidates on their phones, there is a stubborn determination to see through the smoke. People like Ana Luiza Lelis know they will have to look carefully, question everything, and rely on their own instincts. Dona Maria may be a phantom, but the fears she stirred are real, and they have sharpened the senses of a nation. Brazil has been here before, beaten down by lies and bruised by violence, but it has always found its way back to the ballot box. The challenge is to ensure that the next move is guided by human reason, not by the ghost in the machine. The story of Dona Maria is not just a warning; it is also a reminder of what is at stake. When a nation loses the ability to believe its own eyes, it must learn instead to believe in the vigilance of its citizens. In the end, the election will be a test not only of candidates but of the collective will to uphold the fragile, beautiful dream of democracy in an age of manufactured reality.