Abstract: As a regional power in Latin America and a representative of emerging economies, Brazil exhibits a strong identification with the “Global South” in global governance. At the same time, its domestic regulatory framework demonstrates normative dependence on the European Union (EU), reflecting a dual emphasis on both “digital sovereignty” and “de-risking”. Brazil’s approach to global artificial intelligence (AI) governance oscillates among the three major paradigms of the United States (US), China, and the EU, seeking a balanced strategy that asserts leadership while remaining pragmatic. This paper first identifies the overall lack of representation of Latin America in global AI governance, along with the region’s simultaneous exposure to both risks and opportunities. Through an analysis of Brazil’s domestic regulation and global summit participation, this paper further reveals the disjunction between Brazil’s domestic practices and global governance discourse, as well as its triple dependency on the three key players.

Since 2022, generative AI, exemplified by ChatGPT and DeepSeek, has been continuously reshaping global value distribution and geopolitical dynamics. Amid the intensifying US-China technological rivalry, the world has witnessed the ongoing failure of multilateralism in global summits such as Seoul and Paris. At present, regulatory and governance power over AI remains largely in the hands of tech giants, whose profit-driven nature places the Global South under structural exploitation. The global anarchy in AI global governance is calling for more assertive and autonomous actions by nation-state actors, both domestically and internationally. As a regional power in Latin America and a representative of emerging economies, Brazil’s engagement in global AI governance offers a significant point of reference for how Latin American and Global South countries might position themselves and act amid the current wave of technological transformation.

1. AI Global Governance Background

At present, global AI governance exhibits a distinctly realist character, reflecting the principle of great power consensus. Three major governance patterns, respectively from the US, China, and the EU, have dominated the landscape. The US model emphasizes corporate autonomy and is characterized by an “administrative-led” approach, aiming to maintain technological leadership while extending its domestic AI governance paradigm abroad. The EU’s governance framework is rooted in its longstanding regulatory standards on data, privacy, and technology, alongside a traditionally risk-averse economic stance. It prioritizes legislation and regulation negotiated among member states over time, but its pace lags significantly behind the rapid diffusion and sectoral impact of generative AI towards society (Mukherjee et al., 2023). The Chinese model, in contrast, externally emphasizes a Global South narrative and demonstrates strong features of technological catch-up and state-led direction, though it remains rhetorically vague and limited to high-level principles. Domestically, it is marked by government leadership, policy continuity, and collaborative participation from enterprises, with ongoing refinement through “preemptive legislation” (Mok, 2023).

Along the global value chain, the upstream segments of AI infrastructure (energy, telecommunications base, cloud storage, etc.) and foundation model have reflected tech- and capital-intensive patterns, turning the global AI race into a contest primarily among great powers. The DeepSeek-led open-source wave has marked a “Sputnik Moment” in the foundation model domain, prompting many countries outside of China and the US to recognize new opportunities for development. Resource-constrained nations have begun to explore strategic AI applications tailored to specific sectors, thereby reducing their dependence on foreign tech giants and envisioning the possibility of cultivating autonomous AI ecosystems (Yeyati & Guilera, 2025). This shift has also brought the concerns of the Global South into closer alignment with those of the EU. EU member states have expressed intentions to ease digital regulation and promote industrial development, invoking discourses of “digital sovereignty” and “equality” for European industrial renaissance (Mui, 2025; Macron, 2025). 

For Latin America, 2025 represents a critical window of opportunity to shift from the role of a “passive consumer” to that of an active “participant” with data sovereignty and localized technological applications. However, as seen previously with LLaMA3, the open-sourcing of DeepSeek-R1 does not automatically lead to technological democratization. Sergio Amadeu, a Brazilian scholar and former director of the National Institute of Information Technology (ITI), argued that for countries technologically dependent on the US, this is a moment to formulate autonomous development strategies, yet neither open-source adoption nor isolated legislation can address the deeper issue of sovereign infrastructure, and the achievement of digital sovereignty requires a systemic approach (Xiong, 2025). Reliance on foreign AI infrastructure, multinational “algorithmic black boxes”, and unclear data ownership all pose risks that Latin America may be subject to “algorithms designed in other countries and distorted in local contexts” (Yeyati & Guilera, 2025). Moreover, the privatized deployment of powerful models may create additional governance challenges, causing “black boxes” at the implementation level of the policy cycle, and exacerbating Latin America’s existing social crises, like transnational crime, firearms, and drug trafficking.

2. Domestic Governance Practices: Digital Sovereignty & Regulation-Based Pattern

As a regional power in Latin America and an active participant from the Global South in global governance, Brazil’s national development plan for AI and its domestic regulatory framework are embedded in the discourse of “digital sovereignty”. Brazil’s domestic AI development actions mainly follow its Brazilian Artificial Intelligence Plan 2024-2028 (2024). This plan aims to promote innovation and efficiency of AI adoption, especially in the public sector (IDS News, 2024). It includes “Immediate Impact Actions” to solve domain-specific problems in society and economy, and “Structuring Actions” to promote long-term computational capability via the development of the Santos Dummont supercomputer. 

Under a similar regulation-based pattern of the EU, the Brazilian AI development plan focused on encouraging innovation, safeguarding development space, and protecting human rights. Specifically, its framework includes the proposal of a national “sovereign cloud” to protect sensitive data, and the establishment of a national center for AI transparency and trustworthiness. In terms of development plans, Brazil’s development objectives are focused on digital security, social justice, and long-term infrastructure projects, with less attention paid to algorithmic innovation, academic breakthroughs, etc., reflecting a follower or catch-up profile.

On the regulatory side, there is currently no specific AI law in Brazil. AI governance has been mainly placed under the umbrella of data regulation, aligning Brazil with the EU pattern. In 2020, the Lei Geral de Proteção (LGPD), which is similar to the EU’s General Data Protection Regulation (GDPR), was enacted as a localized data protection framework. However, compared to the GDPR, the LGPD places more emphasis on sovereign control of cross-border data flows, local compliance for in-country data processing, as well as the relative independence of legislation and single government ownership (Usercentrics, 2024). In general, the Lula administration has always been wary of transnational technology capital. In August 2024, it issued a ban on X for spreading false information and failing to appoint a legitimate state representative. As a result, the Brazilian government has also attracted Musk’s critique concerning “democracy” and “free speech” (Global Government Affair, 2024). Within the framework of the LGPD, AI governance in Brazil emphasizes transparency, interpretability, and legitimacy, but lacks a specific legal definition of AI and algorithms, regulation of non-personal data, and governance mechanisms for specific AI risks (deep fakes, black boxes, etc.). As a result, while Brazil highlights sovereignty and autonomy in both AI development and regulation, its regulatory model still largely references EU standards, leading to a rule-taker role in AI global governance.

The year 2025 is the last year for the Lula government to push for major policy changes before the elections, making it a crucial period for digital policymaking in Brazil (Covington, 2024). Since 2022, Brazil has been working on the development of the AI Act, following the EU pattern and defining a risk-based approach to regulation. After much bargaining between the government, experts, and stakeholders, the Act was approved by the Senate in December 2024. It is able to apply to all stages of the AI lifecycle, introducing the concept of a “regulatory sandbox” and integrating local priorities (labour rights, environmental sustainability, etc.) with global standards. However, the Act has also been significantly modified under industry pressure, including the exclusion of social media content curation and recommendation algorithms from the high-risk category in order to reduce censorship requirements (Atanasovska & Robeli, 2025). As an extension of domestic governance, Brazil’s engagement in AI global governance may not follow a path of technological catch-up with first-mover regulatory practice to become a “rule maker”. Instead, Brazil’s AI Act is more likely to become a model for the Global South in terms of local adaptation and distributive justice.

3. Global Governance Engagement: Domestic Extension with Global South Identity

In the post-globalization era, negotiations on emerging global governance issues are often discussed through summits. In summit diplomacy, one mode gathers multiple stakeholders to discuss specific issues, mostly in emerging strategic industries. The other one combines head-of-state diplomacy with multilateral institutional platforms, which have the small-grouping and institutionalized characteristics to enhance attention and focus on issues (Odinius, 2021). The outcomes of these summits are often taken to the UN system for reaffirmation and broader consensus-building.

a) Brazil’s Participation in Global AI Summits 2023-2025

The world-first consensus on advanced AI risk and regulation was achieved at the 2023 Bletchley Park AI Safety Summit. This small-scale summit gathered representatives from 28 countries, the EU, the UN, and selected tech giants (OpenAI, Anthropic, Google DeepMind, X, etc.). Although China was the one that was informally invited by the UK for “unacknowledged attendance”, key players in the AI realm still reached an agreement on the Bletchley Declaration (Stacey & Milmo, 2023). Brazil was also invited to attend this summit as the only representative from Latin America, and signed the agreement with the others together. However, participants are not treated equally at this summit. Despite the consensus of the first day’s ministerial meeting, the head-of-state summit on the second day only included the UK and US and their key allies to insist “democratic values” but without democratization of participation. 

This Western-centered summit governance framework continued in Seoul, where Brazil was also present, but with no information about whether it signed any agreements. Brazilian companies were also absent from the signing of the Frontier AI Security Commitment during the summit (Birch & Ozfirat, 2024). It was not until the February 2025 Paris AI Action Summit that this series was expanded from its original mini-summit to a “summit week”. In parallel with the inclusion of the Global South, Brazil’s participation increased. Represented by the Minister of Foreign Affairs, Mauro Vieira, the Brazilian government and most of the participating countries reached a consensus on the inclusiveness and sustainability of AI. At the meeting, Brazil advocated for AI governance to be structured in a way that promotes development and reduces inequality, expressed its support for UN-centered multilateralism and human rights protection, and shared its experience in digital governance (Garcia, 2025).