AI Chatbots Pose Threat to Electoral Integrity in Brazil’s Presidential Race

Brazilian AI Chatbots Defy Electoral Rules Six Months Before Presidential Vote

Six months ahead of Brazil’s presidential election, artificial intelligence chatbots continue to rank and recommend candidates despite new electoral regulations explicitly prohibiting such behavior, an AFP investigation has found.

In January, Carmen Lucia, head of Brazil’s electoral court (TSE), warned that AI chatbots could “contaminate” the October vote in Latin America’s largest democracy. Following these concerns, the TSE implemented strict regulations in March restricting how chatbots may operate during the 2026 election cycle and increasing platform liability for false content.

Under these new rules, AI tools are forbidden from providing recommendations, rankings, or opinions about candidates and political parties, even when directly asked by users. However, tests conducted by AFP in recent weeks revealed that leading AI services—including ChatGPT, Grok, and Gemini—continue to violate these restrictions.

When prompted about the “best candidates for the 2026 elections,” ChatGPT responded with clear preferences, stating: “Honest conclusion. The ‘technically’ best options today: Tarcisio/Zema.” This recommendation referenced São Paulo Governor Tarcisio de Freitas, who has ruled out a presidential bid, and former Minas Gerais Governor Romeu Zema, a potential candidate from the right-wing Novo party.

The chatbots also offered assessments of 80-year-old incumbent President Luiz Inácio Lula da Silva, praising his “vast experience” while criticizing his “advanced age.” Lula, a veteran leftist politician, is seeking an unprecedented fourth term in office.

These ongoing violations raise significant concerns about technology’s potential influence on voter behavior in Brazil, a highly polarized nation where social media and digital platforms play crucial roles in political discourse. Experts warn that chatbot responses are generated through probability calculations based on training data that may contain errors or inherent biases.

“AI systems make recommendations based on the data they were trained on, which inevitably contains biases and factual errors,” said Theo Araujo, director of the Amsterdam School of Communication Research. A study conducted by Araujo during the 2025 Dutch elections found that approximately one in ten voters were likely to use AI chatbots to research candidates.

The TSE’s proactive stance against misinformation isn’t unprecedented. The electoral court previously declared far-right former president Jair Bolsonaro ineligible to run for office after determining he had spread false information about Brazil’s electoral system during the 2022 presidential campaign.

Misinformation concerns extend beyond candidate rankings. In March, AFP’s fact-checking team identified a fake image allegedly showing Flavio Bolsonaro—son of the former president—with Daniel Vorcaro, a businessman under investigation in a major banking fraud scandal. When questioned about this image, X’s AI chatbot Grok incorrectly verified it as authentic and even fabricated a date for the supposed meeting.

When contacted about these violations, OpenAI stated that ChatGPT is “trained not to favor candidates” and that the company continues to refine its models. Google offered a more distanced response, explaining that Gemini generates responses based on user prompts, which do not necessarily reflect the company’s views.

The challenges faced by Brazil’s electoral authorities highlight the global struggle to regulate rapidly evolving AI technologies in political contexts. With six months remaining before Brazilians head to the polls, the effectiveness of the TSE’s regulations remains in question as technology platforms continue to navigate—and at times circumvent—restrictions designed to safeguard electoral integrity in one of the world’s largest democracies.