Artificial intelligence (AI) has gained ground in a new front of the 2026 elections: the mass production of offensive images against women in politics. A survey by the MonitorA project identified signs of AI use in 25% of the images classified as attacks or insults directed at candidates on X.
Another 9% of the cases fell into an area of uncertainty, where researchers could not determine whether there was or not AI manipulation.
The analysis gathered 11,137 images published in 9,393 comments on profiles of 132 candidates. The contents were monitored during two periods of the campaign: between August 16 and 24, at the start of the contest, and from September 2 to 11, after the crisis involving the Supreme Federal Court (STF) and the Master case.
In total, 660 images were classified as offensive against candidates on X.
AI does not necessarily seek to deceive
- The result is striking because the AI usage identified in the study is not concentrated in realistic deepfakes produced to deceive voters;
- Most of the analyzed images consist of caricatures, memes, montages, and fake campaign materials created to humiliate, ridicule, or dehumanize the candidates;
- "These uses of montages and low-quality synthetic images indicate that, in these cases, there is not necessarily an intention to simulate reality." AI appears, above all, as yet another tool for the production and dissemination of offensive content," summarizes the report;
- MonitorA divides offensive content into two categories. Insults are offenses with hostile language, harsher than a critique, but not directly using markers of inequality, such as gender, race, or religion. 269 images were identified in this category;
- Already, the attacks seek to belittle and delegitimize candidates based on characteristics associated with historically marginalized groups, such as gender, race, sexuality, disability, or religion. 391 contents of this type were found;
- According to MonitorA, due to their discriminatory nature, these attacks should be removed by the platforms.
Marina Silva, on the other hand, appears mainly in contents classified as misogynistic, dehumanizing, and ageist. In some images, she is portrayed as an animal or an alien, while others use artificial alterations to change her appearance and make her look older.
Alone, Marina was the target of 17% of all attacks and insults directed at the monitored candidates.
In the overall classification of attacks, misogyny appeared in 99 images. Then came fatphobia, with 55; transphobia, with 46; dehumanization, with 39; and racism, with 27.
Most common types of violence involving AI
- Misogyny: 25,3%;
- Fatphobia: 14,1%;
- Transphobia: 11,8%;
- Dehumanization: 10,0%;
- Racism: 6,9%.
Marina Silva is one of the most offended candidates through AI – Image: Focus Pix/Shutterstock
The offenses against her had two main axes: sexualization, with bikini images and comments implying sexual interest, and questioning her political identity, associating her with a supposed "false conservatism."
Vanessa da Silva, on the other hand, was the target of 30 offensive images. According to the study, 70% contained racist attacks, including representations related to the slave period.
The report establishes a parallel between the two cases: while Julia appears in content with sexualizing offenses, Vanessa is portrayed as enslaved or in publications that associate her with feelings of disgust.
Read more:
- Starlink will be used by TSE in remote areas during the 2026 elections
- STF expands free access to TSE services and fact-checking websites during the elections
- 2026 Elections: searches on Google about electoral fines and sale of beverages spike
MonitorA had already identified violence against female candidates at the beginning of the campaign
The most recent analysis continues the monitoring done by MonitorA during the campaign. In a survey conducted between August 16 and 24, the project identified 1,318 attacks and insults against women on X. During that period, Marina Silva accounted for 32% of the records, followed by Erika Hilton with 23%, and Bia Kicis with 11%.
MonitorA has been tracking patterns of gender-based political violence on digital platforms since 2020. In 2026, the fourth edition of the project brings together InternetLab, Journalism Center and Democracy in Check Institute, which participate in the collection, methodology, analysis and production of the surveys.
TSE created rules for the use of AI in elections
The use of AI during the campaign has also become specifically addressed by the Electoral Justice.
By regulating the 2026 elections, the Superior Electoral Court (TSE) established restrictions on synthetic content produced or altered by AI. Among the rules is the limitation, in the 72 hours before and the 24 hours after the election, of the circulation of new synthetic content that modifies the image, voice or manifestation of candidates and public figures.
The regulation also provides for the responsibility of providers in certain situations involving unidentified synthetic content or that violate other electoral rules. The TSE also prohibited alterations in photographs, videos or other audiovisual records that contain scenes of sex, nudity or pornography, in relation to the protection of the dignity of female and male candidates.
In September, the court also presented tools developed by technology companies to help protect the image and voice of women in politics against improper use of AI.
However, the study by MonitorA shows that the use of technology during the campaign is not limited to the production of content that tries to look real. AI can also be used to accelerate the creation of memes, satires, montages and other offensive materials directed at female candidates.
Rodrigo Mozelli is a journalist graduated from the Universidade Metodista de São Paulo (UMESP) and, currently, is a writer for Olhar Digital.
