Instituto Sergio Surugi de Siqueira - Reaearch Bioethics

A successful first Global Dialogue on AI Governance should deliver outcomes that are both principled and operational—capable of guiding real-world implementation across diverse regulatory, cultural, and resource contexts. First, it should establish a shared baseline of principles for trustworthy AI, with particular emphasis on transparency, accountability, human oversight, and proportionality. These principles must be actionable, not merely aspirational, and adaptable to sector-specific needs such as healthcare, scientific research, and public administration. Second, the Dialogue should produce a practical interoperability roadmap. Fragmented regulatory approaches risk creating uncertainty and inefficiencies, especially for cross-border research and innovation. A common framework that aligns key concepts—such as risk classification, data governance, and auditability—would significantly reduce friction while preserving national sovereignty. Third, it should promote capacity-building commitments, particularly for low- and middle-income countries. Without investment in technical expertise, institutional infrastructure, and ethical review systems, global governance risks deepening existing inequities in access to and control over AI technologies. Fourth, the Dialogue should recognize the urgent need for governance of AI use in scientific research and ethics review. AI systems are already influencing protocol design, data analysis, and ethical decision-making. Clear guidance is needed to ensure transparency of AI use, preservation of human responsibility, and protection of research participants. Finally, success would include the creation of a multi-stakeholder implementation mechanism—not just a declaration, but a living process involving governments, academia, industry, and civil society. This mechanism should support continuous evaluation, iterative improvement, and the sharing of best practices. In short, the Dialogue will be successful if it moves beyond consensus toward coordinated, measurable, and ethically grounded action.

• Safe, secure and trustworthy AI
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Transparency, accountability, and human oversight
• Open-source software, open data and open AI models

I selected these priorities because they address the core conditions for AI to be both ethically legitimate and practically viable across diverse contexts. Safe, secure and trustworthy AI is foundational. Without reliability, robustness, and protection against misuse, AI systems can generate harm at scale, particularly in sensitive domains such as healthcare and scientific research. Trustworthiness is not only a technical requirement, but also a prerequisite for societal acceptance. The social, economic, ethical, cultural, linguistic, and technical implications of AI must be considered in an integrated manner. AI does not operate in a vacuum; it reshapes power dynamics, access to knowledge, and decision-making processes. Ignoring these dimensions risks reinforcing inequalities, marginalizing underrepresented populations, and producing solutions that are technically sound but socially inadequate. Transparency, accountability, and human oversight are essential to preserve responsibility in increasingly automated environments. Users and affected individuals must be able to understand how decisions are made, while clear lines of accountability must be maintained. Human oversight ensures that AI remains a support tool rather than a substitute for ethical judgment, particularly in high-stakes contexts. Finally, open-source software, open data, and open AI models are critical for equity and innovation. Openness enables independent scrutiny, reproducibility, and capacity building-especially in low- and middle-income countries. It helps prevent concentration of power in a few actors and supports the development of locally relevant solutions aligned with diverse regulatory and cultural realities. Together, these priorities form a coherent framework that balances innovation with responsibility, global coordination with local adaptability, and technological advancement with ethical integrity.

Yes. While the listed themes are comprehensive, an important cross-cutting issue that deserves explicit recognition is the governance of AI use within scientific research and ethical review processes. AI is increasingly embedded across the research lifecycle-from protocol design and literature synthesis to data analysis and even the preparation of ethics submissions. However, current governance discussions tend to focus on AI as an object of regulation, rather than as a tool actively shaping how research is conceived, conducted, and evaluated. This creates a regulatory and ethical gap. One emerging issue is the need for clear standards on the disclosure and traceability of AI use in research. Researchers should be required to transparently report where and how AI tools were used, ensuring accountability and enabling proper assessment by ethics committees and reviewers. Another critical aspect is the preservation of human responsibility in AI-assisted ethical decision-making. AI can support ethical analysis, but it must not replace deliberative judgment. Governance frameworks should explicitly reinforce that responsibility for ethical decisions remains with researchers and oversight bodies. Additionally, there is a growing need for AI literacy and capacity building among ethics committees and research institutions. Without this, there is a risk of inconsistent evaluations, overreliance on AI outputs, or unjustified rejection of valid methodologies. Finally, the Dialogue should address the development of domain-specific AI systems trained on ethical and regulatory frameworks. General-purpose models often lack the contextual sensitivity required for research ethics, increasing the risk of misleading or non-compliant guidance. Addressing these issues would strengthen the integrity, transparency, and global coherence of AI governance in science-an area that is rapidly evolving but still insufficiently regulated.

In Brazil—and more broadly in Latin America—the governance gaps in the selected areas are already producing tangible effects, particularly in the research and health sectors. A key challenge is regulatory asymmetry and fragmentation. While new legal frameworks are emerging, their interpretation and implementation remain uneven across institutions. This creates uncertainty for researchers and ethics committees, especially when AI tools are used in protocol design, data analysis, or decision support. The absence of clear, operational guidance on AI use in research ethics leads to inconsistent standards and, at times, either over-restriction or uncritical acceptance of AI-generated outputs. Another significant issue is limited institutional capacity. Many research institutions and ethics committees lack training and technical infrastructure to critically assess AI-assisted research. This increases the risk of opacity, automation bias, and insufficient oversight—particularly in high-stakes studies involving human participants. At the same time, there are important opportunities. Brazil has a strong tradition in research ethics governance and centralized review systems, which can be leveraged to integrate AI governance more coherently. If properly aligned, AI can improve the quality, consistency, and efficiency of ethical review, supporting researchers in clarifying risks, consent processes, and data protection strategies. The emphasis on transparency, accountability, and open models also creates an opportunity to reduce dependency on proprietary systems and foster locally adapted solutions. Open and domain-specific AI tools can support capacity building, promote reproducibility, and enhance trust. Finally, there is a strategic opportunity to position the region as a leader in ethical AI for research, by integrating robust governance with practical, scalable tools that support both innovation and the protection of research participants. In summary, the main challenge is governance capacity and clarity; the main opportunity is to build structured, ethically grounded AI integration in research systems.

The AI Dialogue can play a pivotal role as a convergence platform—not to impose uniform regulation, but to align core concepts, reduce fragmentation, and enable practical cooperation across jurisdictions. First, it can advance regulatory interoperability by fostering a shared understanding of key notions such as risk classification, transparency requirements, auditability, and human oversight. Even without full harmonization, aligning these concepts reduces uncertainty for cross-border research, innovation, and deployment. Second, the Dialogue can function as a mechanism for knowledge exchange and capacity building, particularly supporting low- and middle-income countries. By sharing best practices, tools, and institutional models, it helps prevent a widening global gap in AI governance capabilities. Third, it can promote common standards for the responsible use of AI in scientific research—an area still underdeveloped globally. Establishing guidance on disclosure of AI use, traceability, and preservation of human responsibility would strengthen research integrity and facilitate international collaboration. Fourth, the Dialogue can encourage multi-stakeholder coordination, bringing together governments, academia, industry, and civil society. This is essential to ensure that governance frameworks are both technically informed and socially legitimate. Fifth, it can support the development of open and trusted infrastructures, including open-source tools and evaluation frameworks, which enable transparency, independent scrutiny, and broader participation in AI development. Finally, the Dialogue can act as a continuous process rather than a one-time event, enabling iterative alignment as technologies evolve. By establishing working groups, pilot initiatives, and follow-up mechanisms, it can translate high-level principles into measurable and adaptable actions. Pilot initiatives using domain-specific AI tools for research ethics review can serve as testbeds for international cooperation. In essence, the AI Dialogue can transform international cooperation from abstract consensus into coordinated, practical governance.

The AI Dialogue should build upon existing **normative, regulatory, and technical initiatives**, while focusing on connecting them in a more operational and globally inclusive way. Key references include the **OECD AI Principles**, which provide a widely endorsed foundation for trustworthy AI; **UNESCO's Recommendation on the Ethics of AI**, which emphasizes human rights and cultural diversity; and the evolving regulatory frameworks such as the **European Union AI Act**, which advances risk-based governance. In addition, initiatives like the **Global Partnership on AI (GPAI)** and the **World Health Organization guidance on AI in health** offer valuable sector-specific insights. Despite their strengths, these efforts often operate in parallel, with limited interoperability and uneven global uptake. The added value of the AI Dialogue would be to act as a **bridging and translation layer** between high-level principles and real-world implementation. First, it can promote **conceptual alignment**, reducing fragmentation by harmonizing key terms such as risk, transparency, and accountability across frameworks. Second, it can enable **practical convergence**, by supporting pilot projects, shared evaluation methodologies, and cross-border testing environments—particularly in areas like scientific research and ethical review, where AI is rapidly being integrated but remains under-governed. Third, it can strengthen **inclusion and capacity building**, ensuring that low- and middle-income countries are not merely adopters but active contributors to governance models. Finally, the Dialogue can foster **continuous coordination**, linking existing initiatives into a dynamic ecosystem rather than isolated efforts. In essence, its added value lies not in creating new principles, but in making existing ones **coherent, operational, and globally accessible**.

Governments should provide regulatory perspectives, share policy experiences, and commit to interoperability efforts. Academia can contribute evidence-based analysis, methodological rigor, and independent evaluation of AI systems. Industry should bring technical expertise, implementation insights, and transparency about system design and limitations. Civil society plays a critical role in representing societal interests, identifying risks, and ensuring inclusiveness and accountability. International organizations can facilitate coordination, standard-setting, and capacity building across regions. To be effective, the AI Dialogue should adopt a multi-layered and action-oriented structure: Thematic Working Groups Organized around key priorities (e.g., trustworthiness, transparency, AI in research), these groups should produce concrete outputs such as guidelines, taxonomies, and policy recommendations. Pilot and Testbed Initiatives The Dialogue should support real-world pilot projects to test governance approaches in practice—particularly in sensitive areas like healthcare and scientific research. These pilots can generate evidence on what works and what does not. Regional Hubs and Inclusion Mechanisms Decentralized regional forums can ensure that diverse cultural, legal, and socioeconomic contexts are reflected, avoiding a one-size-fits-all approach. Capacity-Building Tracks Dedicated programs for training regulators, ethics committees, and institutional leaders are essential to reduce global asymmetries in AI governance. Transparency and Continuous Feedback Outputs should be publicly accessible, with mechanisms for iterative review, stakeholder feedback, and periodic updates. Finally, the Dialogue should be designed as a continuous process, not a one-time event—combining global coordination with local implementation, and principles with measurable, practical outcomes.

First, low- and middle-income countries (LMICs) are often insufficiently represented, despite being significantly affected by AI deployment. Their inclusion requires not only invitations, but structural support—funding for participation, regional hubs, and capacity-building programs that enable sustained engagement. Second, practitioners in applied domains, such as healthcare professionals, research ethics committee members, and data stewards, are rarely heard. These actors deal directly with real-world implementation challenges—particularly regarding AI use in scientific research and decision-making—yet governance debates are often dominated by policy and technical perspectives. Including them through practice-oriented working groups and pilot initiatives would ground discussions in operational realities. Third, communities affected by AI systems, especially vulnerable or marginalized populations, are frequently excluded. Their perspectives are essential to identify risks related to bias, discrimination, and unequal access. Mechanisms such as participatory consultations, community panels, and impact assessments should be institutionalized. Fourth, linguistic and cultural diversity remains a barrier. Much of the debate occurs in a limited number of languages and frameworks, which restricts broader participation. Multilingual platforms and culturally adapted materials are necessary to ensure meaningful inclusion. Fifth, open-source and independent research communities are sometimes overshadowed by large institutional and corporate actors. These groups play a key role in transparency, auditing, and innovation, and should be more systematically integrated into governance processes. Inclusion requires moving beyond representation toward enabled participation—providing resources, access, and structured roles in decision-making. A more diverse set of voices will lead to governance frameworks that are not only more equitable, but also more robust and applicable across contexts.

First, policy and governance labs ("co-creation labs") can bring together policymakers, researchers, industry, and civil society to collaboratively design solutions to concrete problems (e.g., AI in healthcare or research ethics). These sessions should produce tangible outputs such as draft guidelines, model policies, or evaluation frameworks. Second, real-world case simulations can be highly effective. Participants could analyze realistic scenarios—such as the use of AI in clinical research or public decision-making—and work through ethical, legal, and technical dilemmas. This format encourages practical reasoning and exposes differences in regulatory and cultural approaches. Third, pilot showcases and testbed reviews should be included. Organizations can present ongoing AI governance pilots, followed by structured peer review and feedback. This creates a learning loop between theory and implementation. Fourth, multi-stakeholder roundtables with defined outcomes can replace open-ended discussions. Each session should aim to produce a short, actionable output (e.g., a set of agreed principles, risks, or recommendations), ensuring that dialogue translates into progress. Fifth, digital participation platforms can expand inclusiveness. Structured online consultations, asynchronous contributions, and multilingual engagement tools allow broader global participation beyond those physically उपस्थित. Sixth, reverse panels or "role-switch" dialogues—where, for example, policymakers take the role of affected communities or developers respond as regulators—can surface blind spots and foster empathy across perspectives. The Dialogue should include iterative feedback cycles, where outputs are revisited, tested, and refined over time. This ensures that engagement is not only dynamic, but also cumulative and impact-oriented.

At the policy level, frameworks such as the OECD AI Principles and UNESCO's Recommendation on the Ethics of AI establish important normative foundations, while regulatory models such as the European Union risk-based AI framework demonstrate how to translate these into enforceable structures. However, a critical frontier in AI governance lies in how these principles are operationalized in real-world decision-making contexts. One emerging approach is the development of domain-specific, AI-assisted governance platforms. For example, ETHEXPRESS-maintained by the Instituto Sergio Surugi de Siqueira-is an AI-supported system designed to assist researchers and institutions in navigating ethical and regulatory requirements in scientific research. It integrates structured ethical reasoning, alignment with applicable norms, and strict human oversight, ensuring that AI acts as a support tool rather than a decision-maker. This type of platform illustrates a shift from abstract governance to embedded, real-time ethical guidance, reducing ambiguity and improving consistency in high-stakes environments. In parallel, mandatory disclosure policies for AI use in research, open-source audit ecosystems, and controlled pilot environments ("regulatory sandboxes") are advancing transparency, accountability, and iterative learning. These developments suggest that effective AI governance will not be achieved by regulation alone, but through a combination of normative frameworks, enforceable rules, and purpose-built systems that operationalize ethical principles at the point of use. Scalable, domain-adapted platforms represent a key next step in making AI governance both practical and globally transferable.