Electoral Prosecutors Probe AI Chatbots for Ranking Candidates
Brazil’s Electoral Prosecutor’s Office opened an investigation into major AI chat platforms after civil-society complaints and studies found systems ranking or recommending presidential pre-candidates in ways that appear to breach Superior Electoral Court rules, Folha and other outlets reported in early September 2026. Prosecutors sought information from the TSE about meetings with AI platforms and notified companies including OpenAI, Google, xAI, DeepSeek, Anthropic and Perplexity to file preliminary responses, following research that flagged candidate favouritism in a large share of tested answers.
Filed under Policy and dated September 20, 2026, this AI4Brazil briefing treats the probe as Brazilian election-integrity AI news distinct from Redata data-centre tax stories. TSE resolutions already bar chatbots from ranking, recommending or favouring candidates, yet August and September tests kept surfacing preferential answers even after some firms filed compliance plans. The dispute sits months before voters go to the polls, when synthetic media and ranking bias can move faster than takedown desks.
Why it matters: Brazilian voters increasingly ask chatbots who is ahead. Unchecked ranking can tilt information diets—but only enforceable logs, geo-filters and penalties will change product behaviour.
What it means in practice
Brazilian campaign counsel and platform counsel should inventory which AI surfaces answer election queries in Portuguese; demand documented refusal modes for ranking prompts; assign an owner for weekly red-team tests; run time-boxed audits against TSE resolution text; and prefer contracts that force rapid model patches. Connect the probe to the pending AI framework vote and the data watchdog’s AI centre.
Caveats come first. Prosecutorial notices are not convictions; model behaviour changes daily; and studies sample limited prompts. AI4Brazil therefore presents the chatbot investigation as directional policy context until TSE findings and platform remediation reports appear.
What to watch next: TSE hearing outcomes; published compliance patches; and how Manacá Portuguese models sit beside foreign chatbots in civic use. Readers can continue on the AI4Brazil homepage, or browse the Newsroom for additional briefings.
Bottom line: treat this update as orientation, not instruction. Brazilian election AI enforcement is moving from paper rules to active probes and remains early. Organizations that benefit most will test Portuguese ranking refusals, keep humans on campaign decisions, and refuse to confuse a preliminary notice with finished compliance.