AI-Generated Political Influencers Emerge in Brazil's Elections
According to Bloomberg, a new wave of AI-generated political influencers has emerged in Brazil's presidential election campaign. These synthetic personas speak in the accents, skin tones, and complaint lists of target voters, creating "people like you" as mass-producible campaign units. As Latin America's largest economy is already highly connected, this round expands influencer politics from real accounts to virtual ones.
A typical character named Dona Maria sits in front of a microphone lamenting rising supermarket bills and accusing Lula of diplomatic incompetence in his dealings with U.S. President Trump. The videos have accumulated millions of views. Election monitoring organizations have identified 18 AI avatars used for political evaluation between January 2025 and April 2026, with 11 of them, about 60%, not disclosing their synthetic identity; 78% of the content contains false or distorted statements about politicians, democratic institutions, or public policies. Dona Maria is portrayed as a Black woman and has been observed posting over 400 times since August 2025, targeting the president and Supreme Court justices. Opposing personas like "Mr. Market" specifically attack the right wing.
A mid-year survey by AtlasIntel showed that over 150 million Brazilians are online, with 63% following political news from influencers, and over 70% of respondents believe these faces will influence voting. Free broadcast television slots for the election began airing on August 28, with the first round expected in early October. Current President Lula faces off against Liberal Party candidate Flávio Bolsonaro. Former President Jair Bolsonaro, who is banned from campaigning and under house arrest, cannot personally solicit votes, but his AI stand-in spoke at his son's campaign launch, claiming to be a simulation calling for support for his chosen successor.
The Superior Electoral Court has decided this year to prohibit the creation, replacement, or alteration of the images and voices of living, deceased, or fictional characters using synthetic audio and video for electoral purposes, and requires explicit labeling of synthetic propaganda. In practice, supporter accounts continue to post unmarked content. A newspaper monitored over 100,000 fake accounts from both sides five days after the election began, capturing 25 videos that generated or altered the appearance of candidates, 22 of which were unmarked, totaling millions of interactions. The Labor Party has filed a lawsuit regarding the generated video of Bolsonaro at the conference, with the court set to hear the case on September 1, potentially establishing a definition for deepfakes: whether it can mislead voters into believing something actually happened. The vice president has proposed a line based on "sufficiently creating a false sense of reality," regardless of which side the content supports.
The sellers are studios and freelance creators who can cover unlimited events with a single face; the buyers are camps looking to bypass legal restrictions on candidates while cheaply saturating information flows. Funding has shifted from traditional TV ads and real influencer commissions to computing power, prompts, and replicable personas. Beneficiaries include political brands that can still appear despite being silenced, as well as recommendation systems on platforms that push content based on completion rates. The pressure is on electoral courts that must legislate simultaneously for "fictional voters" and "simulated candidates," as well as voters who cannot distinguish between a neighbor and a model. The driving force is the upcoming October voting and the existing gaps in prohibitions, rather than a single scandal.
Source: Public Information
ABAB AI Insight
Brazil is not suddenly being attacked by artificial intelligence; it is an upgrade of the mobilization seen on WhatsApp and YouTube since 2018. Back then, it was about real people sharing chains and church or convoy groups; now, the "neighborly aunt" has been transformed from an unscalable labor force into a finely-tuned asset. Dona Maria doesn't sleep, doesn't demand payment, and doesn't contradict herself in cross-examinations, yet she can post daily updates along three emotional tracks: inflation, diplomacy, and judges. Creators can be ride-hailing drivers in Rio, with production costs close to the cost of a single generation, while distribution costs are borne by platform recommendations. This follows the same logic as Jair Bolsonaro using a stand-in after being deprived of public political expression: the law can control bodies, but not models.
The capital path is shifting identity rents from celebrity accounts to template libraries. Real influencers face risks of breach, changing their stance, or being bought out; synthetic personas turn risks into version management. What the camps want is not a viral hit, but a matrix that covers different state accents, skin tones, and complaints about grocery baskets. Once labeling rules are only written in resolutions and not incorporated into recommendation algorithms, the completion rates of unmarked videos will systematically be higher than compliant ones. The definition battle at the Superior Electoral Court on September 1 is essentially about which comes first: the "simulated candidate" or the "simulated voter" being cut off: the former involves the agency of banned leaders, while the latter affects the supply curve of daily information flows.
Comparable cases include cloned candidate calls in the 2024 U.S. local elections, synthesized hosts in Indian dialects, and digital avatars of deceased politicians in South Korean elections. Brazil is at a turning point from expansion to control: the connected population and influencer dependency are already saturated, while regulation is still chasing this year's virtual voters with last year's deepfake provisions. The industry position is not about technical demonstrations, but rewriting the organizational advantages of the last social media election into model advantages.
Structurally, it is a layering of technological substitution and regulatory changes. Substitution occurs in the labor force for persuasion: one shoot is replaced by one inference. Regulatory changes happen as the legal definition of "person" fails to keep pace with the dissemination definition of "looking like a person." The mechanism is: electoral laws hold subjects accountable, while recommendation systems reward similarity; whoever can mass-produce similarity will harvest attention in the lag of enforcement. Courts can remove a video from the conference but cannot eliminate eighteen unmarked aunts speaking simultaneously across five platforms.
ABAB News · Cognitive Laws
- The law can control bodies but not replicable faces.
- The cheapest voter is one who doesn't need to sleep.
- Labeling is written in resolutions, while traffic is written in models.