AI-Generated Political Avatars Evade Disclosure in Brazilian Social Media
An Observatory for Elections study identified 18 AI-generated avatars used to spread political disinformation on Brazilian social media between January 2025 and April 2026. Reportedly 61% of the synthetic profiles failed to disclose their artificial nature, with 78% containing misleading allegations about politicians including President Lula, with the case of 'Dona Maria' becoming the most prominent example.
What happened
A research study by Observatório das Eleições, conducted by Data Privacy Brasil and Aláfia Lab, identified 18 AI-generated avatars used to comment on politics on Brazilian social media between January 2025 and April 2026. The study found that 61% of the analyzed publications did not indicate that content was produced by artificial intelligence.
The artificial avatars appeared as supposed voters, commentators, influencers, and popular leaders. Fourteen of the 18 cases mapped, equivalent to 78% of the total, contained misleading allegations about politicians or democratic institutions, with targets including President Luiz Inácio Lula da Silva, former president Jair Bolsonaro, and Supreme Court ministers including Alexandre de Moraes, Cármen Lúcia, and Luís Roberto Barroso.
In many cases, the artificial origin could only be identified after technical analysis, which revealed signs such as resolution failures, proportion differences, and roboticized characteristics in images and audio. The Tribunal Superior Eleitoral (TSE) had established rules requiring explicit notification when material was created or manipulated with AI, but the study showed lack of compliance.
Reported impact
The most emblematic case involved 'Dona Maria,' an artificial avatar depicted as a Black elderly woman who published over 400 videos between 2025 and 2026, criticizing the federal government and attracting significant reach for spreading political disinformation.
The avatar's reach prompted a legal action filed with the TSE by PT, PV, and PCdoB parties requesting suspension of profiles linked to the character. The phenomenon also produced a countereffect when left-wing pages published their own pro-government version of 'Dona Maria' with the same physical characteristics but favorable discourse toward President Lula.
Index analysis
Among the seven cases with some type of warning or indication, the alerts did not follow a uniform pattern. Some warnings were added automatically by the platforms themselves, while others appeared through watermarks from the tools used or hashtags included in publications.
Governance controls
The Tribunal Superior Eleitoral established rules requiring explicit notification when AI was used to create or manipulate election-related material, with identification of the technology used in the process. However, the research demonstrated that these requirements were not being followed by most creators of AI avatars engaging in political discourse.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.