Brazil's Superior Electoral Court is staring down a concrete test of whether any national judiciary can compel a platform giant to neutralize its own algorithms during an election. Sleeping Giants Brasil and CTRL+Z filed new evidence Monday showing that searches for "brasil" on WhatsApp and Instagram surface images of Sen. Flávio Bolsonaro — his ticket number, his running mate — while Lula appears only as text. The same GIF search reproduced the result across three separate accounts, suggesting a systemic ranking pattern rather than personalization noise. The derogatory asymmetry is more pointed. A WhatsApp GIF search for "ladrão" (thief) returned a caricature of President Lula in prison stripes and a message calling Workers' Party support a mental illness. On Instagram, the same query surfaced "Vote 13" — Lula's party number. Hashtag ordering tells the same story: searching "lula" offered "#lulaladrão" (Lula thief) as the top suggestion, while the critical "#bolsonaronuncamais" (Bolsonaro never again) was buried at the bottom of Bolsonaro-related results. The organizations also tested Meta AI on WhatsApp, asking two users to pick a number between 1 and 25 with no electoral context. One got 13 (Lula's party number), the other got 22 (Bolsonaro's). On its own, the test is thin — two data points from a 25-number range prove nothing statistically. But the organizations are not trying to prove bias in a lab. They are building a legal record of persistent asymmetry across multiple surfaces — search, recommendations, hashtags, and AI responses — that together describe a pattern. The court is not treating this gently. Justice Marco Aurélio Bellizze had already ordered Meta to apply identical criteria to both campaigns. The full court upheld that order Friday. On Sunday, Bellizze issued another 24-hour compliance deadline and warned the daily fine could double from R$500,000 (~$99,000) to R$1 million (~$198,000). Meta's response so far: blame technical failures, offer no specifics, and go silent when Folha de S.Paulo called for comment on the new filings. The structural question is whether algorithmic neutrality is even possible when recommendation systems are optimized for engagement, not fairness. Meta's products rank content by predicted interaction — and polarizing, negative, or emotionally charged content reliably generates more engagement. A system that surfaces "Lula thief" GIFs faster than neutral content may not be "biased" in the sense of a human thumb on the scale; it may simply be doing exactly what it was designed to do. The court's order demands equal treatment, but Meta's architecture rewards unequal virality. Meta has disclosed nothing about the criteria driving its search rankings, hashtag ordering, or Meta AI responses in Brazil. The company's refusal to explain its ranking logic — even under direct court order — is the clearest signal of where power actually sits. A sovereign court can issue fines; a platform can absorb them as a cost of doing business if the alternative is opening its algorithmic black box to judicial scrutiny. The protest planned at Meta's São Paulo headquarters — titled "Meta, Hands Off Our Election" — adds public pressure to a legal process that may ultimately test whether any democracy can enforce algorithmic accountability against a company with the resources and incentive to stall indefinitely.
