Electoral Courts and the Regulation of Social Media in Brazil: An Analysis of the Anti-Democratic Disinformation Regulation during the 2022 Elections ()
1. Introduction
The construction and creation of political narratives are not new phenomena. However, starting in the 2010s and intensifying in the 2020s, they entered a heightened level of scrutiny due to the possibility of rapid and large-scale dissemination via social media. This environment triggered profound political shifts resulting from the conjunction of factors such as the self-regulation of big tech, the interests of these technology giants, the workings of social media algorithms, the algorithmic targeting of content, the actions and interests of content creators, and the lack of transparency and data. This scenario became conducive to disinformation, understood as the deliberate dissemination of inaccurate information intended to deceive and to misinformation, which refers to the unintentional spread of inaccurate information but shared in good faith. Fabricated fake news emerged as influential political agents, shaping elections and placing democracy at risk with the assistance of social media.
Considering these considerations and establishing a well-defined thematic scope, this study examines the necessity for specific legislation under Brazilian law to regulate the activities of technology giants regarding transparency and network responsibility in the face of large-scale fake news dissemination, drawing on the discussion of how such practices may influence the democratic status of the Brazilian State. A jurimetric approach is employed to evaluate the situation in Brazil based on the actions of the Superior Electoral Court (TSE) and the Regional Electoral Courts (TERs) when adjudicating cases involving anti-democratic fake news disseminated on social media during the electoral process. Given the political context under review and the objectives presented here, the following hypotheses were proposed: 1) The number of decisions regarding anti-democratic fake news on social media has increased with each electoral cycle; 2) Most of decisions’ foundations rely on explicit legal texts directly related to the issue of fake news; 3) The specific legal bases concerning fake news used in the decisions exhibit low variation.
2. Methodology
This study encompasses both fundamental and applied research, adopts a propositional objective, and integrates bibliographic, documentary, and jurimetric approaches. It employs a mixed-method design that combines quantitative and qualitative techniques, employs content analysis, and incorporates an instrument-validation procedure.
Bibliographic research was conducted to build the theoretical framework and identify the main issues surrounding the topic, while documentary research concentrated on Brazilian legislation related to democracy. This approach also supported the selection of databases for the jurimetric research, which guided data collection and pattern analysis, initially using quantitative methods.
Regarding the methodological approach, the study adopts a mixed design that combines quantitative and qualitative methods in the data mining and analysis phases. ChatGPT was applied for content summarization and synthesis to optimize the examination of large volumes of text, and content analysis was applied to classify the decisions concurrently with an instrument-validation procedure to verify the suitability of the configured ChatGPT-based agent.
This study is both theoretical and applied in nature. From the theoretical perspective, it seeks to build knowledge about replicating the adopted procedures, particularly regarding the Chat GPT configuration created agent. On the applied side, with a propositional objective, it endeavors to apply the knowledge produced to practical contexts, especially in developing specific legislation on fake news.
2.1. Database and Temporal Scope
The Democratic Rule of Law, which underpins the Brazilian State (Article 1, Federal Constitution of 1988, CF/88), is upheld by an institutional framework that encompasses multiple entities. Although the Federal Supreme Court (STF) is the guardian of the Constitution (Article 102, CF/88) (Brasil, 1988) and thus responsible for examining matters related to maintaining the democratic structure, this study—given its focus on the dissemination of anti-democratic fake news during the electoral process—has concentrated on the Electoral Justice database. The goal was to gather data from the Superior Electoral Court (TSE) and the Regional Electoral Courts (TREs) since the Brazilian Electoral Justice, whose structure and functions are defined by the Federal Constitution (Brasil, 1988) and the Electoral Code, Law No. 4.737/1965 (Brasil, 1965), is attributed with organizing, conducting, and overseeing the electoral process to ensure legitimacy and transparency of elections.
A filter was applied to limit the searches to judgments under the rationale that, being collegial decisions, they have a greater impact on setting precedents. Consequently, this set of decisions has a higher capacity to enhance the relevance, accuracy, and reliability of the results. By concentrating on this type of decision, the intention is to ensure greater data consistency, optimize the investigation toward the most pertinent information, and ensure result validity by working with a homogeneous set of information.
Regarding the temporal scope, although the creation and dissemination of fake news and the use of historical narratives for political purposes are not new phenomena, their large-scale spread via social media began in the second half of the 2010s and worsened in the 2020s. In line with this study’s focus, filters were applied based on the classification of results referencing electoral cycles, excluding decisions not tied to election years. Finally, given the recent nature of the topic, no restrictive temporal filter was imposed on the initial period of analysis, and the results were limited to the 2022 electoral cycle, as the 2024 election remains too recent for thorough examination.
2.2. Descriptors and Boolean Operators
Given the available search options in the Electoral Justice database, data collection was divided into two stages: the first gathered data from the Superior Electoral Court (TSE), and the second collected data from the Regional Electoral Courts (TREs) concurrently. In view of the database’s specialized nature, only one search was conducted for each stage, applying the same descriptors and Boolean operators: (“fake news” OU “notícia falsa” OU “notícias falsas”) + “redes sociais”.
Quotation marks were used to designate complete search terms (descriptors). Thus, for example, the search would retrieve results in which the descriptor “fake news” appeared as a single phrase rather than yielding results containing the words “fake” or “news” individually. The same guideline applies to synonyms, incorporated in the search using the OU (OR) Boolean operator, which indicates meaning that any element within the parentheses can be used interchangeably. The Boolean operator + was employed to signify a necessary connection between descriptors, ensuring that at least one descriptor within the parentheses appears in the search, while the descriptors linked by + must also be present.
2.3. Data Analysis Procedure
The research results were examined both quantitatively and by reviewing the content of the obtained documents. The evaluation of their relevance to the study was conducted after organizing and mining the data in Excel and using a GPT tool specifically developed for summarizing information.
The data collected were organized in Excel spreadsheets, which allowed for the following actions:
Identifying duplicate entries using Excel’s highlighting function;
Searching for duplicates through comparison;
Categorizing the overall results;
Determining the suitability of each result for the study based on the document’s content;
Analyzing the content of those results was considered relevant to the study.
To perform faster and more effective analysis of the data regarding the hypotheses, thereby optimizing the examination of large textual volumes, an artificial intelligence tool available on OpenAI’s ChatGPT platform was employed. A specialized GPT tool configuration was used to generate summaries, as illustrated in Table 1 below.
The capabilities were disabled to prevent access to external data beyond what was provided—in the case of Web Browsing—and to exclude functions unrelated to the intended use of the tool in the case of the DALL-E Image Generation feature. No actions enabling the GPT tool to retrieve information or perform tasks outside of ChatGPT were configured, likewise aiming to prevent interference from external information not contained in the supplied documents. The summaries generated by the GPT tool were reviewed individually, and additional questions were posed to supplement the information summarized and organized by the configured tool.
Table 1. Programming a GPT-based tool for summarizing higher court decisions.
Model version used |
ChatGPT 4o |
Category |
Research & Analysis |
Description |
Summarize procedural documents from Brazilian higher courts. |
Instructions |
Summarize procedural documents using simple language according to the following specifications: OBJECTIVE: Seek content related to the widespread dissemination on social media of false or fraudulent news (fake news) that undermines Brazil’s democratic order. CLASSIFICATION—The documents must be categorized into one of the following groups: 2.1. Decisions related to the electoral period 2.2. Decisions related to fake news 2.3. Decisions related to participation in anti-democratic acts 2.4. Decisions related to Covid-19 2.5. Decisions related to freedom of the press 2.6. Decisions related to parliamentary immunity 2.7. Decisions related to incitement to violence 2.8. Other subjects IDENTIFICATION STAGES—After classification, the following steps must be performed: 3.1. Case identification 3.2. Type of record (judgment, single-judge decision, presidential decision, bulletin, etc.) 3.3. Reporting, reviewing, and sentencing judges (if applicable) 3.4. Parties involved 3.5. Dates of trial and publication SUMMARY STAGES—Present the following points in a concise manner: 4.1. Summary of the request addressed by the document 4.2. Legal grounds of the request, with particular emphasis on the applicable legal provisions 4.3. Decision 4.4. Basis of the ruling in the specialized literature 4.5. Basis of the ruling in the legal framework, emphasizing the legal provisions employed //GENERAL GUIDELINES: A new analysis must be carried out for each document provided without allowing previous analyses to affect the results of a new analysis. Questions asked previously about documents provided in the past should not serve as a basis for analyzing newly provided documents. |
Capabilities |
Code interpreter and data analysis |
Denied capabilities |
Web browsing DALL-E image generation |
Actions |
None |
3. Development
The pivotal milestone of the cognitive revolution essential to the flourishing of
human species—and which culminated in the supremacy of Homo sapiens sapiens—was the cerebral shift that enabled the joint operation of multiple intelligences, occurring approximately 70.000 years ago (Mithen, 1998). In a groundbreaking manner, the capacities to imagine, create fiction, and narrate—combined with social necessities, already reflecting a political inclination—distinguished the human species from the others cohabiting the planet. Consequently, creating and guiding narratives can be understood as phenomena intrinsic to human existence, deployed according to the interests of groups and individuals. In the 21st century, especially from the 2010s onward, social media propelled political narratives to a new level, allowing for rapid and large-scale dissemination.
Inequality relations observed throughout history as part of the collective living dynamic, exemplified by Hegel’s interpretation of historical dialectics through the master-slave dialectic (Hegel, 2003), have likewise undergone modifications. Among various contributing factors, technological advancement has amplified these issues through globalization, enabling more powerful control and exercise of hegemonic authority, exemplified by the more aggressive role of the culture industry and the information regime. According to Han (2022), this information regime is a form of domination in which information and its processing by algorithms and artificial intelligence decisively determine social, economic, and political processes, extending beyond bodily control to the realm of data and information. Nevertheless, this does not rule out interpreting the information regime in tandem with Foucault’s conceptualization of biopower, presenting ever more complex problems (Foucault, 2009).
The phenomena stemming from the interplay between globalization and capitalism have transformed consumption practices, group and individual relations, and political processes. The culture industry influences the design of consumer goods, the attribution of value, and the targeting of consumer audiences toward adopting or acquiring ideas, ideals, goods, and services. Adorno and Horkheimer (2002) raise the concern that society operates under specific directives that curtail freedoms, with the result being the imposition of selected goods, content, discourses, identities, and more, guided by economic and political interests. Thus, the freedom to choose ideology, which always reflects economic coercion, reveals itself in every domain as the freedom to choose what is invariably the same thing.
Hegemonic forces that assert and reaffirm themselves throughout historical and cultural processes, the importation and adoption of epistemological, legal, political, and social models, the market logic under which the entire world is forced to pass through the filter of the culture industry (Adorno & Horkheimer, 2002), the growing power of the media over intellectual life (Lipovetsky & Serroy, 2008), the massification of culture and cultural hybridity (Canclini, 2012), heterogeneous identities (Bauman, 2005) that nevertheless undergo fragmentation or pluralization (Hall, 1996), the direction that social media platforms impose on societies, and the role and strategies of big techs are all factors that generate and exacerbate social issues.
With big data and the widespread implementation of artificial intelligence, new forms of domination have emerged. According to Han (2023), artificial intelligence gains access to our desires and inclinations of which we are not aware. This constitutes a capture of behavior at the unconscious level to shape how society functions (Han, 2020). For O’Neil (2017), opacity, scale, and harm are the three elements that turn mathematical models—used to organize vast datasets and to program algorithms—into weapons of destruction.
Fisher (2022) highlights the urgent need to regulate social media because, without oversight, the currency of exchange between users and platforms is screen time. Engagement, therefore, drives the dissemination of content to ever-larger audiences, regardless of the nature of the information being circulated. The author warns about the operation of algorithms, pointing to a lack of control over false and even criminal information, which can be boosted organically or through sponsorship in monetizable posts. The chief concern lies in how big techs manage and take responsibility for events incited by their algorithms, such as anti-vax campaigns, hate speech, school attacks, election outcomes, and so forth.
In Brazil, there have been instances of democratic questioning, historical denialism, the suspension of memory policies, and the creation of erasure policies directed at facts that remain central to agendas advocating for memory and truth. Other actions similarly characterize the production and steering of political narratives. Nonetheless, changes in social behavior are observable in the public spheres of various countries, featuring phenomena such as chronic disinformation, political radicalization, and growing belief in conspiracy theories. According to Gray, Bounegru, and Venturini (2020), these do not necessarily point to isolated failures within the systems but rather to structural shifts tied to the unregulated expansion of platformization.
With the purpose of regulating freedom, transparency, and accountability on the internet, Bill No. 2630/2020 (Brasil, 2020), popularly known as the Fake News Bill, is undergoing a bicameral review process, with one phase already concluded in the Federal Senate (Senado Federal, 2020). Despite its title, its objective is to hold platforms accountable for conduct already defined as offenses in the Brazilian Penal Code, require big techs to disclose periodic reports to society and ensure these companies act during emergencies or particularly concerning situations. However, it is essential to note the slow legislative progress since 2020, the rejection of placing this measure under urgent consideration in 2022 (Câmara dos Deputados, 2022), and the lack of any clear regulatory timeline.
The self-regulation of big techs, or more properly, bug platforms fosters a scenario that primarily serves corporate interests, including those tied to lack of transparency, which in turn shields these companies from accountability for numerous social and political outcomes. Consequently, the continuity of this structure relies on a scarcity of data—data that would otherwise depend on external research but remain partly obscured by missing information, such as details on how algorithms operate. These algorithms collect personal data to gauge users’ interests and determine the delivery of content. For this study, no data or research was found indicating the volume of fake news circulating on social media. As a result, to achieve a valid understanding of the problem, additional data relevant to the dissemination of fake news on social platforms was sought.
Fake news spreads 70% faster than real news. Although this finding stems from a study that only assessed posts on X (formerly Twitter), specifically tweets written in English, similar outcomes emerged when the research was replicated in other countries (Vosoughi, Roy, & Aral, 2018). Beyond the speed that typifies this phenomenon, the Brazilian context shows worse indicators regarding frequency of occurrence, the ability to detect false information, and overall trust in social media. According to research by Poynter et al. (2022), 44% of Brazilians report receiving fake news daily. Additionally, a study by the Organization for Economic Cooperation and Development (OECD, 2024), employing a gamified methodology, determined that Brazilians perform the worst at identifying false news, achieving a 54% accuracy rate. Furthermore, while the global average of those who have high trust in social media stands at 9%, in Brazil, more than 20% exhibit a high level of trust.
4. Results
For this study, the parameterization of the Brazilian Electoral Courts’ actions regarding the dissemination of anti-democratic fake news was carried out to observe the variation in relevant demands over time and examine the legal grounds cited in their decisions. It is assumed that these data may support a better understanding of the phenomenon’s progression and help identify possible gaps stemming from the absence of specific legislation, thereby facilitating the proposal of practical measures that can uphold the democratic framework. Data organization and analysis followed the steps below:
1) Stage 1: Organization and elimination of repeated items, constituting the results initially considered for thematic relevance;
2) Stage 2: Classification of the outcomes from Stage 1 and analysis of thematic relevance;
3) Stage 3: Quantitative and qualitative analysis of results with confirmed thematic relevance for hypothesis verification.
4.1. Data Mining
The TSE jurisprudence search returned a total of 35 results, classified by the Electoral Justice database as referring to the 2018 (2 judgments), 2020 (17 judgments), and 2022 (16 judgments) elections. Only 2 duplicated items were identified (classified as pertaining to the 2020 election), resulting in 34 entries initially considered for thematic relevance. Of these, 14 were deemed pertinent to the subject matter and proceeded to the subsequent stages of hypothesis verification.
The simultaneous jurisprudence search across the TREs produced 27 results, classified by the Electoral Justice database as referring to the 2016 (2 judgments), 2020 (10 judgments), and 2022 (15 judgments) elections. No duplications were detected, so all 27 entries were analyzed for thematic relevance, of which 19 were considered pertinent, and proceeded to the hypothesis verification stage.
The initial data mining results from Stages 1 and 2 are shown in Table 2.
Table 2. Initial mining results.
|
Initially considered for thematic adequacy analysis |
Decisions excluded due to lack of topical relevance |
Results with thematic relevance |
TSE |
34 |
20 |
14 |
TREs |
27 |
8 |
19 |
4.2. Data Classification
The decisions initially considered for thematic relevance were grouped into four categories: Illicit Electoral Advertising for rulings on early campaigning, non-compliance with electoral regulations, or other anomalies; Sharing of False Polls; Fake News on Social Media; and Other for rulings not classifiable into categories containing more than five elements.
The decisions were classified based on an individual content analysis of each judgment with respect to the issue under review, employing a qualitative content-analysis methodology. Content analysis of the rulings revealed that the case records lacked sufficient information on the posts’ dissemination modalities—organic, pseudo-organic, or sponsored—thereby precluding the study from differentiating among these categories.
Multi-thematic judicial decisions were identified within the classification groups “Fake news on social media” and “Unlawful electoral propaganda”. In such cases, the decision was classified according to the component most relevant to the study—the dissemination of fake news on social media—even when the primary issue adjudicated was unlawful electoral propaganda, and the fake news component was treated only as an ancillary matter.
The classifications were distributed in Figure 1 and Table 3.
Figure 1. Categorizing the overall results.
Table 3. Classification of results based on thematic adequacy.
|
Illicit Electoral Advertising |
Fake news on social media |
Others |
Total |
TSE |
0 |
14 |
0 |
|
TREs |
5 |
13 |
1 |
33 |
4.3. Data Analysis
Stage 3 of the results mining process enabled both the verification of the hypotheses and the instrument validation carried out on the configured GPT’s use and performance in content analysis. The main findings are described below.
Hypothesis 1: The number of decisions regarding anti-democratic fake news on social media has increased with each electoral cycle.
The thematically relevant claims produced outcomes starting in the 2016 electoral cycle, with minimal numerical variation when compared to 2018. A similar pattern was observed for the 2020 claims directed to the Superior Electoral Court (TSE). However, regarding the demands brought to the Regional Electoral Courts (TREs), the year 2020 showed a notable upturn, with a considerable increase. This upward trend persisted in 2022 and was accompanied by a similar rise in cases presented to the TSE, which surged in comparison to the previous election cycle, as illustrated in Figure 2.
Based on the data, it was possible to observe, especially from the 2020 and 2022 electoral cycles onward, that the volume of anti-democratic fake news on social media is on the rise, confirming the projected hypothesis.
Hypothesis 2: Most of the decisions’ foundations rely on explicit legal texts directly related to the issue of fake news.
Considering the absence of specific legislation covering freedom, transparency, and responsibility on the internet under Brazilian law—but considering the possibility of drawing on other legal frameworks to address this gap, such as the General Data Protection Law (Brasil, 2018)—it was hypothesized that most decisions would rely on legal texts expressly related to the issue of fake news. Upon examining the applicable provisions, however, out of 32 decisions, only 9 were based on legal provisions directly addressing this topic, thereby refuting the hypothesis, as shown in Figure 3.
Figure 2. Demands by electoral cycle.
Figure 3. Distribution of decisions regarding the use of legal grounds with thematic specificity.
Figure 4. Use of provisions from TSE Resolution No. 23.610/2029 in decisions.
Hypothesis 3: The specific legal bases concerning fake news used in the decisions exhibit low variation.
Upon analyzing the legal grounds cited to justify the decisions, variations were noted in the provisions applied to similar matters. Among the nine decisions that provided specific rationales referencing fake news, however, no variation emerged in the legal bases employed. Rather, Articles 9-A, 17(§1-B), and 27(§1) of TSE Resolution No. 23.610/2019 (Brasil, 2019) were repeatedly invoked, with the specific proportions shown in Figure 4.
4.4. Instrument Validation: GPT Agent
The configured GPT agent, used to summarize the data from the documents, was applied freely by uploading documents as necessary. Thus, the total number of analyzed results does not match the number of times the documents were inserted into the instrument. One explanation for this discrepancy is the need to verify data even after classification, address any doubts about a given case, and divide the data mining process into stages.
Throughout the process, the tool repeatedly altered the aesthetic format of the output, although the precise number of occurrences was not recorded, given that it did not influence the information itself but simply reorganized headings and topics, leaving the overall order largely intact. However, the absence of consistent patterns is significant, as the results had to be compiled in spreadsheets and were then edited manually to ensure aesthetic uniformity, contrary to the objective of time optimization.
On three occasions, formatting changes compromised the information provided, resulting in requests to revert to the previous format or pay closer attention to the programming. The problematic format changes did not occur abruptly between outputs but rather appeared as small modifications over time, ultimately leading to output formats that failed to convey all the information defined in the programming.
Among the documents analyzed, eight could not be processed by the tool due to an unknown error, and no pattern was identified to explain the issue. Four of these documents came from the TSE jurisprudence research, while the other four were from the simultaneous TRE jurisprudence search. Of these eight, seven were successfully copied and pasted into the chat box for analysis without major complications, whereas one document proved too lengthy to be processed in its entirety. Consequently, this decision was analyzed individually, with GPT tool support limited to the summary and digest text available in the Electoral Justice database.
Although the configured GPT tool was used solely for summarizing and expediting the analysis of large text volumes, it is recognized that using GPT tools may pose challenges for scientific reproducibility since artificial intelligence can behave differently depending on user interaction. Therefore, when employing the programming described here, it is advisable to implement necessary adaptations based on other researchers’ needs and to adopt an instrument validation procedure to document behavior patterns and assess any deviations.
5. Propositional Conclusion
Building on the objective of examining the need for specific legislation within Brazilian law to regulate the activities of big tech companies, particularly regarding transparency and accountability on social media in the context of widespread dissemination of anti-democratic fake news capable of undermining the democratic framework of the Brazilian State, this research proposed three hypotheses. These were tested through an analysis of the role played by the Superior Electoral Court and the Regional Electoral Courts in matters involving anti-democratic fake news during the electoral process. The research, including jurimetric analysis through both qualitative and quantitative lenses, supports the following conclusions and conjectures.
The lack of large-scale data on fake news circulating on social media—stemming directly from the absence of transparency around big techs’ operations—hampers a more comprehensive assessment of the issue. This limitation affects the depth and precision of scientific evaluations and underscores the need for suitable evaluative parameters based on available data, such as indicators of social behavior changes (which could be derived from digital marketing tools not employed in this study) and the legal demands and responses of the Judiciary, as examined here through a jurimetric approach.
The observed increase in the number of decisions related to anti-democratic fake news on social media with each electoral cycle objectively demonstrates the escalating severity of the problem and the professional use of such resources. On a more abstract level, it could also signal a growing familiarity with the subject, though this familiarity does not necessarily denote a higher level of appropriate technical knowledge.
The facts that most legal rationales in these decisions do not rely on legislation directly addressing fake news and that the few that do are limited to TSE Resolution No. 23.610/2019, with little variation in the underlying arguments, point to the absence of a specific, foundational legal framework. More abstractly, this may also suggest a lack of technical expertise regarding the topic.
Assessing the mounting seriousness of the problem through these indicators—seemingly exponential, as shown by the numerical difference between the 2016 and 2018 elections compared to 2020 and 2022—objectively highlights the need for regulatory measures governing big techs in Brazil. At the same time, the legal bases used in the relevant decisions further underscore the necessity of a dedicated regulatory framework, not only for upholding freedom, responsibility, and transparency in online networks but also for addressing the dissemination of fake news. This issue is likely to become even more critical given both the increasing professional application of these resources and advances in technology, illustrated by deepfakes—synthetic images and/or human voices generated by artificial intelligence.
Acknowledgements
C. A. U. thanks the University of Fortaleza, and the Fundação Cearense de Apoio ao Desenvolvimento Científico e Tecnológico—Funcap (Ceará Foundation for the Support of Scientific and Technological Development) for the financial support provided through a doctoral scholarship.