MAIN ARTICLE
A penalty for noncompliance, not a verdict on bias
On November 10, the Superior Electoral Court, known as the TSE, unanimously imposed the penalty on Facebook Serviços Online do Brasil Ltda. The judgment centered not on which candidate benefited from digital campaigning, but on the company’s response to court directives during a particularly sensitive stage of the election. The TSE said some requests were answered late, partially or in ways that prevented reliable verification. When the remaining campaign is measured in days, information delivered after voting ends cannot perform the same public-interest function as timely disclosure.
The amount comprises three days at 500,000 reais, from October 17 through 19, and six days at 1 million reais, from October 20 through 25. Reporting justice Marco Aurélio Bellizze criticized fragmented responses, information supplied in foreign languages and broad assurances that did not demonstrate substantial compliance. The distinction matters: a sanction for failing to obey court orders is not proof that a company deliberately manipulated political choices. Conflating the two would weaken rather than strengthen public understanding of the ruling.
The court also maintained confidentiality over certain technical submissions while allowing the complainant access to compliance documents, except protected descriptions of AI filters and synthetic-content detection methods. That approach exposes a genuine tension. Businesses have legitimate trade secrets, yet secrecy cannot become an all-purpose obstacle to legal oversight when their systems influence electoral debate.
From post visibility to artificial intelligence
Lula’s Brasil Pronto Pra Mais coalition argued that posts supporting the incumbent faced restrictions while material associated with Flávio Bolsonaro enjoyed more favorable reach. Its filing also raised concerns about allegedly inauthentic accounts, identification of AI-generated material and oversight of political advertising.
These are related but distinct questions: organic recommendations, paid amplification and chatbot answers involve different technical processes and require different methods of evaluation.
In mid-October, Brazilian reporting described complaints that Meta’s AI tools rejected some requests involving Lula while accepting comparable requests involving his rival. The campaign portrayed the difference as a sign of systematic discrimination. Yet an individual tool malfunction, or a disparity in video views, does not establish its own cause. Software failures, publishing strategies, repeat viewing and user behavior may all shape observed results. Scrutiny therefore demands more precise evidence than screenshots or aggregate impressions alone.
The TSE requested data capable of independent assessment rather than general promises of neutrality. A system may apply formally identical rules while producing unequal effects; equally, an unequal outcome does not necessarily reveal political intent. Distinguishing these possibilities is essential if democratic safeguards are to avoid becoming punishment based on suspicion rather than proof.
Meta’s rebuttal and the problem of proof
Meta rejects the bias allegations and maintains that it applies the same rules to every candidate, party and political viewpoint. In a November 6 explanation, the company argued that some external comparisons overlooked differences in posting frequency and repeated video views. It said registered Instagram accounts belonging to Lula reached more people overall than their Bolsonaro counterparts during the analyzed campaign period. Meta also cited broader technical problems, anti-fraud controls and compliance requirements rather than selective suppression.
Those figures are relevant, but they do not resolve every question. Aggregate reach across a campaign does not necessarily answer what happened within a particular product, on a given day or under a specific judicial instruction. The TSE recognized technical failures while saying that, at the preliminary stage, it lacked grounds to conclude that Meta had intentionally distorted the race. A finding of deliberate interference would require persuasive evidence of intent in appropriate proceedings.
Agência Brasil reports that the substantive dispute can continue even after the procedural fine. Readers should keep three categories separate: what one political campaign alleged, what the court established about noncompliance, and what remains unproven about purposeful election interference. Clear distinctions protect both the integrity of public scrutiny and the fairness owed to those accused.
The vote is over, but oversight is not
Lula ultimately won the runoff by more than 4.4 million votes after trailing in the first round, according to the Associated Press. Organizing online and on the streets helped shape the contest. Yet the outcome cannot by itself settle questions about how political information moved across digital networks before election day. A winning campaign may still raise legitimate questions, just as an accusation by a winning campaign still requires proof.
Social platforms now provide spaces for political speech and act as systems that rank, filter and distribute that speech. Sensible oversight therefore needs enforceable response deadlines, auditable information, consistent rules and meaningful appeals. Excessive regulation can threaten expression, while unexamined algorithmic power can make competition less transparent. Neither danger disappears because an election has produced a definitive result.
The fine will not answer every allegation about Brazil’s 2026 campaign. Its broader significance lies in establishing a standard for information infrastructure: neutrality should be demonstrable, not merely declared. Public confidence emerges neither from a technology company’s assurances nor from partisan suspicion alone, but from credible evidence, lawful procedures and independent judgment.
















