On March 2, 2026, Brazil’s Superior Electoral Court adopted Resolution No. 23,755, which strengthened electoral rules on AI and platforms. For the October 4, 2026 election, AI system providers cannot recommend candidacies, and certain new synthetic content is banned from 72 hours before to 24 hours after polling day. The court frames certain uses of AI and supplier obligations without treating them as campaign financing violations.
The essentials
- The Superior Electoral Court (TSE) bans the distribution of new synthetic content in the 72 hours before the October 4, 2026 election and the 24 hours following it.
- Since TSE Resolution No. 23,755/2026, providers of applications offering AI systems cannot rank, recommend, suggest, or prioritize candidacies; the standard does not classify this ban as a campaign financing rule.
- Brazil built this framework after the 2022 experience, when disinformation circulated widely, though the TSE already had a normative basis to order platforms to remove content and sanction violations.
- National regulation can hardly constrain global technological infrastructure, and rule enforcement remains uncertain if platforms ignore it outside their priority markets.
2022 changed the game
Brazil’s October 2022 presidential election provided a brutal case study. Synthetic content targeting candidates circulated on WhatsApp and YouTube in significant numbers that fact-checking teams could only partially cover. Telegram groups coordinated distribution before platforms had even identified the content as problematic. The TSE had tools to sanction candidates and parties. The 2024 framework gave the TSE direct regulatory levers over certain uses and results of recommendation systems, without conferring general control over all algorithms.
The 2022 result was tight: Lula beat Bolsonaro by 1.8 percentage points. Brazilian electoral rules regulated human emitters, not automated amplifiers. Platforms benefited from limited liability. Since 2024, TSE electoral rules have imposed active duties and defined responsibility on them. Recommendation systems are subject to precise obligations within the Brazilian electoral framework.
The TSE drew lessons between 2022 and 2026 by publishing three successive series of resolutions, refining at each step the definition of synthetic content, platform obligations, and compliance timelines.
The new rules and their concrete obligations
The resolution adopted in March 2026 establishes two distinct constraints.
The first concerns synthetic content. Any image, video, or audio generated or substantially modified by AI must be labeled as such throughout the campaign. Within this window, certain new synthetic content related to a candidate or public figure is prohibited. The temporal restriction concerns only new synthetic content, without preventing removal of old illegal content under other provisions. Article 9-B requires synthetic content labeling; its § 3-A bans, within the electoral window, their publication, republication, and paid promotion when they use the image, voice, or manifestation of the persons targeted.
The second constraint is more structural. The ban targets AI system functionalities offered by relevant providers. Recommendation AI cannot, during the electoral period, constitute a vector of implicit promotion. This obligation applies according to the functionalities actually offered by each relevant provider. WhatsApp is subject to obligations applicable to its public distribution features, particularly its channels, while the applicability of certain obligations depends on functionalities actually offered.
The sanction for non-compliance is suspension of access to these services on Brazilian territory. The TSE already has a rapid decision mechanism that can produce injunctions within hours.
This system places technical governance at the same level as party financing rules. A party exceeding its spending cap faces electoral sanction. An AI system provider whose service ranks, recommends, suggests, or prioritizes a candidacy violates electoral rules as such, without being treated as a campaign financing violation, but with effects potentially far exceeding traditional advertising purchases.
Platform responsibility, a blind spot in global electoral law
Most democracies built their electoral law around a central idea: rules apply to political actors. Parties, candidates, donors, audiovisual media subject to licensing: these entities are named, regulated, and accountable. Digital platforms long benefited from a passive host status that exempted them from editorial responsibility.
This model was coherent when platforms merely stored content. It fractured when their algorithms began deciding who sees what, based on engagement signals. The recommendation algorithm amplifies certain content and buries others according to internal criteria that electoral regulators do not control. This asymmetry is what Brazil is attempting to correct.
The European Union took a parallel path with the Digital Services Act, which entered into force in 2023, requiring very large platforms to assess systemic risks during elections and implement proportionate mitigation measures adapted to identified risks. The DSA does not impose categorical prohibition on recommending a candidate but conditions compliance on actual risk assessment and effectiveness of implemented measures. On this point, European legislation discussed in this article on digitizing public services illustrates a similar tension: regulating global private infrastructure from a national or regional framework produces blind spots.
The Brazilian approach stands out through targeted bans and verifiable compliance obligations. It combines specific restrictions on AI system providers with compliance plans, indicators, and audits, without imposing a blanket obligation on all recommendation algorithms. An AI system provider that ranks, recommends, suggests, or prioritizes a candidacy would not comply with Brazilian rules.
Platforms facing regulation they no longer control
The effectiveness of this system depends on a condition the TSE does not entirely control: platform cooperation. Recent history suggests this cooperation is conditional.
During the 2020 US elections, Twitter and Facebook accepted reducing circulation of certain contested content under political pressure. In 2022 and 2023, several of these same platforms shifted their policies in the opposite direction, arguing against excess censorship. X eliminated most of its trust and safety teams. Meta reduced investments in moderating political content outside the United States.
In 2024, the Supreme Court ordered X’s suspension for failing to comply with judicial decisions and lacking a legal representative, not the TSE for removing disinformation accounts. X eventually yielded after several weeks, both to maintain access to a market of 215 million users and to prevent its Brazilian leadership from facing personal prosecution. This episode established a precedent: the threat of suspension is credible, and platforms took it seriously.
For 2026, nine platforms submitted compliance plans to the TSE. Plans must describe structures, procedures, channels, and timelines applicable, notably for processing electoral justice requests. The agreement commits technical teams to concrete processes.
Independent verification remains unresolved. The TSE can observe after the fact that content was amplified by an algorithm. It is far more difficult to audit in real time the recommendation systems of platforms whose source code is proprietary.
Lessons from the Brazilian model beyond its borders
Brazil is not the first country to regulate AI during elections. India imposed synthetic content labeling requirements for 2024 elections. The African Union published non-binding guidelines. Several US states adopted deepfake electoral laws. But the Brazilian framework stands out because it combines three rarely united elements: diligence, transparency, and compliance obligations on recommendation systems, a temporal window freezing new synthetic content, and a credible sanction mechanism grounded partly in prior decisions.
International IDEA, the intergovernmental organization monitoring global electoral integrity, published in 2025 an analysis of electoral integrity ecosystems in emerging democracies, noting that Brazil had developed one of the most sophisticated institutional architectures to address digital risks. What makes this sophistication remarkable is that it was built under real pressure, not in a laboratory: the TSE legislated on a problem it had experienced, with the limitations that implies and the priorities it clarifies.
The model remains improvable on one central point: algorithmic transparency. Verification does not rest solely on platform good faith: the TSE provides for documentary mechanisms and audits, despite practical limitations. An independent technical audit framework, with real access rights to recommendation systems, would be the logical next step. Several digital law researchers, notably at the Getulio Vargas Foundation in São Paulo, are working on audit protocols that could provide this additional verification layer without requiring access to complete source code.
Who controls political visibility controls more than advertising
The Brazilian system goes beyond the October 4, 2026 election. It raises a structural question: who makes the rules when political visibility infrastructure is owned by global private companies operating under multiple jurisdictions.
A political party can buy targeted advertising on a platform, and this amount is disclosed. But if that platform’s recommendation algorithm organically amplifies content generating emotional engagement, and if this content statistically favors certain political profiles, the effect can exceed purchased advertising, without ever appearing in a campaign financing register.
This imbalance is a measurement problem. Recommendation algorithms produce differentiated effects depending on content types and user profiles, but available tools to quantify these effects have limits for meeting electoral law requirements. The same algorithmic transparency deficit that arises in the workplace, as analysis of algorithmic surveillance systems shows in another context, appears here with direct democratic stakes.
The TSE imposes targeted obligations on relevant application providers, including those offering generative AI or recommendation functionalities, without indiscriminately submitting all algorithms to the same regime. This pragmatic choice delimits applicability but creates verifiable obligations through compliance plans, indicators, and audits. It at least has the merit of naming the problem in law and placing responsibility where it actually lies: on technical architectures, not merely human emitters.
The challenge for coming years will be whether this framework can be made verifiable. Several trajectories are possible. In the first, platforms cooperate sufficiently that the 2026 electoral period unfolds without major incident, validating the model and prompting other countries to adopt it—Argentina, Colombia, and Mexico all have major elections by 2027 and are following Brazil’s experience. In the second, a blatant violation during the campaign exposes verification limits, forcing either escalation of audit capacity or political weakening of rules under platform pressure.
The determining variable will likely be the TSE’s capacity to build independent technical monitoring infrastructure. Human and technical resources needed to monitor in real time the recommendation systems of five or six major platforms over seventy-two hours are substantial. The difficulties coordinating technical capacity between public institutions Europe experiences in other domains arise here acutely: technical regulation requires technical expertise few electoral authorities possess internally.
Brazil is building in 2026 a precedent on the nature of digital responsibility in democracy. The question posed by the TSE will remain timely long after October 4: at what threshold of influence does a recommendation algorithm become a political actor, and what institutional architecture would enable regulating it as such.
Sources
- Superior Electoral Court of Brazil, Electoral Calendar and AI Regulation for 2026 Elections: https://www.tse.jus.br/comunicacao/noticias/2026/Marco/tse-aprova-calendario-eleitoral-e-regulamenta-uso-de-ia-nas-eleicoes-2026
- International IDEA, Between Resilience and Fragility: Electoral Integrity Ecosystems: https://www.idea.int/publications/catalogue/entre-la-resiliencia-y-la-fragilidad-los-ecosistemas-de-integridad-electoral?lang=es
- European Union Digital Services Act Regulation, entered into force in 2023, official text available on EUR-Lex.
- Getulio Vargas Foundation (FGV), works of the Center for Technology and Society on digital electoral integrity in Brazil.
