For months, a quiet but intense drama has been unfolding in Brazil, a nation famously passionate about its politics and deeply connected online. It centers on the upcoming 2026 presidential election, not just with the familiar human candidates but with a new, digital player: Artificial Intelligence. Imagine a world where you casually ask a computer program, “Hey, who should I vote for?” and it actually gives you a recommendation. That’s precisely the scenario Brazil’s electoral court, the TSE, is trying to prevent, seeing AI chatbots as a potential source of “contamination” that could sway voters and disrupt the democratic process. Justice Carmen Lucia, the head of the TSE, has been vocal about these concerns, recognizing the immense power these digital voices hold, especially in a country as polarized and hyper-connected as Brazil.
The TSE isn’t known for shying away from tough decisions. They’ve already set a global precedent by declaring former far-right president Jair Bolsonaro ineligible to run for office due to spreading misinformation during the 2022 elections. This strong stance against disinformation now extends to AI. In March, they rolled out new rules explicitly forbidding AI tools from offering any kind of political advice – no recommendations, no rankings, no opinions on candidates or parties, even if a user directly asks for it. They also increased the accountability of platforms for any false content generated by these AIs. On paper, it sounds like a robust shield against digital interference.
But here’s where the plot thickens and the shield shows its cracks. Just weeks after these strict new regulations were put in place, independent tests conducted by AFP revealed a startling truth: some of the leading AI chatbots were still openly flouting the rules. When asked who the “best candidates” for 2026 would be, ChatGPT, Grok (X’s AI), and Gemini all chimed in with their opinions. ChatGPT, for instance, suggested São Paulo’s Tarcisio de Freitas (who has actually ruled out a presidential bid) and former Minas Gerais governor Romeu Zema, a potential candidate for the right-wing Novo party, calling them the “technically’ best options today.” President Luiz Inacio Lula da Silva, an 80-year-old veteran leftist seeking a fourth term, was ranked between second and fifth. While the bots praised his “vast experience,” they also, rather controversially, critiqued his “advanced age.”
This isn’t just about a few rogue algorithms; it’s about the potential for technology to subtly or overtly influence millions of votes with information that might be incorrect or, even worse, biased. Theo Araujo, a director at the Amsterdam School of Communication Research, highlights a crucial point: chatbot responses aren’t born from impartial wisdom. They’re generated based on probabilities derived from the vast datasets they’re trained on. These datasets, created by humans, inevitably contain errors or biases. Araujo’s research from the 2025 Dutch elections showed that a significant portion of the electorate – one in ten people – were already turning to AI chatbots for information about candidates. Imagine that scale in a country like Brazil. The implications are profound: if a large number of voters are getting their political insights from these flawed digital sources, the very fabric of democratic discourse could be compromised.
The concern isn’t purely theoretical; there are already concrete examples. In March, AFP’s fact-checking team debunked an image circulating online that falsely showed Flavio Bolsonaro, another prominent political figure, with Daniel Vorcaro, a businessman under investigation in a major banking fraud scandal. Yet, when asked about this image, Grok, X’s AI chatbot, confidently declared it to be real and even provided a specific date for the alleged meeting. This kind of digital fabrication, amplified by a supposedly authoritative AI, can quickly erode trust and spread false narratives like wildfire, especially in a country where political scandals are often met with fervent public debate.
The tech giants behind these chatbots have responded, but their explanations haven’t fully assuaged concerns. OpenAI, the creator of ChatGPT, stated that its model is “trained not to favor candidates” and that it’s continuously refining its algorithms. Google, responsible for Gemini, remarked that its AI generates responses based on user prompts and that these don’t necessarily reflect the company’s views. While these statements indicate an awareness of the problem, they also underscore the inherent challenge: controlling the unpredictable output of powerful AI models that operate on complex, opaque systems. As Brazil heads into its 2026 election, the struggle to reconcile the promise of AI with the imperative of democratic integrity continues, making the digital landscape as crucial a battleground as any traditional campaign trail.