Semantix

The first Global Dialogue on AI Governance represents a historic opportunity — and its relevance will be measured by the concrete, measurable, and human-centered actions it produces. Sovereignty must be a central pillar. Governing AI means governing data, context, and history. This requires a careful balance between national autonomy and global cooperation, avoiding technological dependency and the concentration of power among few actors. The active inclusion of the Global South is not merely a matter of representation: it is a prerequisite for just distribution. Developing countries supply AI systems with data, digital labor, and cultural contexts — yet have historically been excluded from decisions about how these systems are designed and monetized. Equity in AI begins with recognizing that the Global South already contributes to it — and must have proportional voice in shaping its direction. Recognizing the limits of self-regulation is equally essential. Accountability cannot be delegated exclusively to companies. The dialogue's success depends on building independent mechanisms for oversight, auditing, and clear accountability for AI decisions and their long-term impacts. Without this, well-intentioned principles become statements without consequence. Governance must also ensure access, inclusion, and mitigation of labor market impacts. AI cannot deepen existing inequalities; it must foster capacity-building and fair distribution of opportunities. These pillars are necessary — but insufficient without addressing a more fundamental question: AI must be guided by human ethics, not treated as an independent ethical agent. An AI cannot be more ethical than the values and decisions of those who build it. Separating "AI ethics" from "human ethics" risks diluting the responsibility of those who design and feed these systems. Governing AI is, above all, governing human behaviors, incentives, and decisions — especially in systems that accumulate memory and reinforce patterns over time. The success of this dialogue will depend on ensuring that responsibility remains where it has always been: in the hands of those who create, regulate, and decide.

• Safe, secure and trustworthy AI
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Interoperability of governance approaches
• Transparency, accountability, and human oversight

The four selected priorities reflect an integrated, human-centered vision for the future of AI governance. Safety, security, and trustworthiness are foundational conditions for any AI system to be legitimately adopted - without them, all other progress loses its grounding. The social, economic, ethical, cultural, and linguistic implications recognize that AI is not merely a technical phenomenon: it reshapes labor, power, identity, and access, demanding a multidimensional approach that includes especially the contexts of the Global South, historically underrepresented in these discussions. The interoperability of governance approaches is essential to avoid regulatory fragmentation and ensure that countries at different levels of development can cooperate and build common standards without surrendering their sovereignty. Transparency, accountability, and human oversight are at the core of everything - and here lies the most critical and least debated point: the risk is not only in AI decisions, but in how AI shapes human decisions. Systems with persistent memory accumulate patterns, reinforce biases, and silently frame the perception and judgment of those who use them. AI does not replace human decision-making - it precedes, frames, and quietly steers it. Transparency, therefore, is not merely a technical principle: it is an ethical and social imperative. When people do not know how a system was trained or what patterns it perpetuates, their decision-making autonomy is compromised without their awareness - affecting individuals, institutions, and democracies. Governing AI is, above all, governing human behaviors - those at the origin of these systems and those shaped by them over time. Without real transparency, there is no ethics possible - and without ethics rooted in human behavior, AI governance will always remain incomplete.

Yes. Three cross-cutting and emerging issues deserve attention and remain underrepresented in the AI governance debate. Memory, continuity, and system identity. Most governance frameworks treat AI as point-in-time decision systems. But systems with persistent memory accumulate context over time, creating profiles, patterns, and narratives that influence future decisions - both by machines and by the humans who use them. Governing AI memory is a distinct and urgent challenge, involving privacy, manipulation, and human autonomy far more deeply than current debates acknowledge. Regulatory capture as a systemic risk. There is a growing risk that governance processes themselves are captured by the actors they are meant to regulate. Companies with greater technical capacity and resources tend to dominate consultations, define vocabularies, and shape agendas. Without explicit mechanisms to guard against this bias, AI governance risks legitimizing the very concentration of power it should be designed to counter. AI's impact on collective cognition and democracy. Beyond individual effects, AI systems at scale affect how entire societies form opinions, process information, and make collective decisions. This impact on social cognition and democratic processes is still addressed in fragmented ways - across disinformation, algorithmic bias, and polarization - without an integrated approach that recognizes AI as society's epistemic infrastructure. These three dimensions - memory, regulatory capture, and collective cognition - are transversal to all listed themes and require explicit treatment. Ignoring them means governing the surface while leaving the deepest risks unaddressed.

Brazil holds a unique position in the global AI governance debate: continental scale, one of the world's largest volumes of personal data, and a numerically connected population — but one still marked by functional technological illiteracy. Having access to devices and platforms is not the same as understanding how these systems work, what they collect, or how the decisions they make affect the lives of those who use them. This distinction is ignored in the regulatory debate — and that is precisely where it fails. The central challenge is not merely the absence of regulation — it is the risk of poorly calibrated regulation. The LGPD, inspired by Europe's GDPR, already exposed this problem: a framework built on Global North realities, applied to a country marked by deep social inequality, regional diversity, and asymmetric digital access, protects unevenly — and therefore fails to protect everyone as it should. The AI bill risks repeating this mistake. Systems that accumulate memory about historically marginalized populations do not merely reflect inequalities — they crystallize them. Regulation built without grounding in Brazilian reality may be rigid enough to create friction for innovation, yet insufficiently robust to protect those who need it most. The worst of both worlds. This is compounded by the state's limited technical capacity to enforce what it regulates, legislative haste, and the disproportionate influence of external agendas on a process that should be sovereign. The real opportunity lies in slowing down to get it right. Universities, the technology sector, and civil society must occupy this space — before it is filled by corporate interests or models imported without adaptation to Brazilian reality. You cannot protect those who do not know they need protection. Effective governance begins with literacy — and Brazil still has much to build on that foundation.

The Global Dialogue on AI Governance can drive the creation of an international framework based on shared responsibility, going beyond technical principles to incorporate respect for each country's ethical and moral dimensions. This framework should be grounded in a common set of principles — such as transparency, safety, and accountability — while recognizing that moral values and cultural contexts are not universal. Rather than imposing a single model, the dialogue should enable convergence across diverse perspectives, fostering alignment without erasing local identities. Moreover, international cooperation must take an active role in reducing global asymmetries. Developed countries should act as enablers of sovereignty for developing nations, promoting knowledge transfer, capacity building, and equitable access to technology. In this context, AI governance goes beyond regulation and becomes a lever for global development, where responsibility means not only mitigating risks but also distributing opportunities. Ultimately, this model depends on a core principle: responsibility is human, contextual, and non-transferable.

The Global Dialogue on AI Governance should build on existing initiatives to avoid duplication and accelerate impact, acting as a true orchestrator of global cooperation. Key mechanisms to leverage include multilateral initiatives such as UNESCO, with its ethical AI guidelines; the OECD, whose principles are widely adopted; and the G20, which drives strategic global discussions. In addition, regulatory frameworks like the EU AI Act and technical standards from the ISO provide concrete foundations for standardization and interoperability. Partnerships with the private sector and technical communities are also essential, including organizations such as the Partnership on AI, which contribute real-world practices and evidence. However, the Dialogue's added value lies in going beyond connection — acting as an integrator and catalyst for progress. This includes aligning principles with practice through measurable mechanisms; ensuring meaningful inclusion of the Global South in decision-making; and reinforcing that AI governance must be centered on human responsibility, not just technology. Furthermore, the Dialogue can advance underexplored areas such as AI memory governance, which is critical to mitigating the persistence of bias and concentration of power over time. In summary, its role is to connect, integrate, and evolve — transforming fragmented efforts into a more coherent, inclusive, and effective global governance model.

Stakeholders must contribute complementarily to an effective AI Dialogue, ensuring multi-stakeholder, inclusive, and technically grounded governance. Governments define policies and accountability; international organizations foster convergence; the private sector brings practical implementation; academia provides evidence; and civil society ensures inclusion and rights protection. However, the Dialogue must go beyond abstract principles. It should advance toward technical AI governance, covering the full system lifecycle: models, data, memory, and evaluation mechanisms (evals). AI memory illustrates this challenge: it not only stores information but accumulates patterns and shapes future decisions, potentially reinforcing bias and inequality. Yet focusing only on memory is insufficient. It is the interaction between: • how AI decides (models), • what it learns (data and memory), • and how it is evaluated (evals), that determines real-world impact. If evaluation mechanisms are weak or biased, they can legitimize harmful outcomes. Therefore, governing AI means governing the entire technical system and its validation criteria, recognizing that these elements are ultimately defined by humans.

Global discussions on AI governance still face representation asymmetries, undermining both legitimacy and effectiveness. One major barrier is language. The dominance of a single language limits participation from many countries and communities. Making the dialogue natively multilingual, using AI as an enabler, is key to expanding access and inclusion. Voices from the Global South, vulnerable communities, invisible AI workers, and technical professionals outside major hubs remain underrepresented, despite being directly impacted and contributing to these systems. There is also a gap in addressing the cognitive and social impact of AI, especially in systems with memory that shape decisions over time. To address this, inclusion must go beyond symbolic participation, through: • multilingualism by design • active participation with decision power • funding and capacity building • accessible formats Diversity is not optional — it is essential for effective and equitable AI governance.

An effective and innovative format for the AI Dialogue is the "Guided Co-Creation Cycle", combining active participation, experimentation, and structured decision-making. 1. Immersion and alignment (multilingual and inclusive) Short sessions with data, real cases, and dilemmas, supported by real-time AI translation to ensure global access and shared understanding. 2. Practical labs (applied co-creation) Multistakeholder groups work on concrete challenges — such as AI governance, memory, and model evaluation — using sandbox dynamics to test solutions in practice. 3. Validation and commitment (accountability) Proposals are collectively evaluated, prioritized, and turned into measurable commitments with clear ownership, metrics, and timelines. This cycle can be continuous, with digital interactions between sessions, enabling ongoing collaboration and iteration. Differential: it transforms dialogue into a living process of co-creation, validation, and accountability, linking discussion to real-world implementation.

Effective AI governance is emerging through practices that combine regulation, transparency, and human accountability. Key examples include risk-based regulation, which prioritizes high-impact systems; independent audits and continuous model evaluations to ensure accountability; and governance of AI memory, controlling how data is stored and reused over time. Additionally, regulatory sandboxes enable safe experimentation, while open infrastructures and global collaboration help reduce inequalities between countries. Finally, AI literacy is essential for society to understand, participate in, and oversee these systems. Summary: governing AI means structuring human responsibility across the entire lifecycle - from data to decision.