UNED, Spain

A successful Global Dialogue on AI Governance should not be measured solely by consensus-building, but by its capacity to reconfigure the normative grammar of AI governance. Three outcomes are critical. First, the establishment of a shared baseline grounded in international human rights law, explicitly recognising emerging dimensions such as cognitive liberty, mental privacy, and psychological integrity. This requires moving beyond a narrow risk-based paradigm toward a rights-based constitutional approach capable of addressing forms of non-coercive yet structurally pervasive influence, including behavioural manipulation and cognitive surveillance. Second, the articulation of interoperable governance principles that do not flatten normative diversity but instead enable coordination across jurisdictions while respecting plural epistemologies, particularly those from the Global South. Interoperability must be understood not merely as technical alignment, but as normative translation across asymmetrical power contexts. Third, the creation of institutional follow-up mechanisms that ensure continuity, accountability, and epistemic inclusivity. Without this, the Dialogue risks becoming a performative exercise. A multi-stakeholder observatory or iterative review process could anchor long-term governance. Ultimately, success lies in whether the Dialogue can shift the debate from "how to regulate AI systems" to how to govern the socio-technical infrastructures that shape cognition, agency, and democratic life.

• Protection and promotion of human rights
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Transparency, accountability, and human oversight
• Interoperability of governance approaches

These priorities reflect the need to address AI governance as a constitutional and socio-technical challenge, rather than a purely technical or regulatory issue. The protection of human rights is foundational, but must be expanded to include emerging cognitive dimensions. AI systems increasingly operate at the level of attention, perception, and decision-making, raising concerns about cognitive liberty and autonomy that are not yet fully captured by existing legal frameworks. Transparency, accountability, and human oversight are essential to prevent the diffusion of responsibility characteristic of complex AI systems. However, transparency should not be reduced to technical explainability alone; it must include institutional and epistemic transparency, enabling meaningful democratic scrutiny. The inclusion of social, ethical, and cultural implications is necessary to counter the false neutrality of AI systems. These technologies embed values, biases, and power relations, often reproducing structural inequalities along gender, racial, and geopolitical lines. A critical and intersectional perspective is therefore indispensable. Finally, interoperability of governance approaches is key in a fragmented regulatory landscape. Yet interoperability must be approached cautiously, ensuring that it does not become a mechanism for normative homogenisation dominated by technologically advanced actors, but rather a space for plural and equitable coordination.

A central emerging issue insufficiently captured by current thematic frameworks is the rise of cognitive governance, understood as the regulation and modulation of human behaviour through AI-driven environments. While existing discussions focus on safety, fairness, and accountability, they often overlook how AI systems increasingly function as architectures of influence, shaping attention, preferences, and decision-making processes at scale. This raises concerns not only about individual rights but about the collective conditions of democratic agency. In this context, emerging concepts such as cognitive liberty point to the need to reconsider how existing rights frameworks address the protection of mental autonomy, particularly in environments characterised by pervasive data extraction, behavioural prediction, and algorithmic mediation. Relatedly, the concept of cognitive surveillance, the extraction and analysis of behavioural and neural data to predict or influence mental states, requires urgent conceptual and regulatory development. Current data protection regimes are ill-equipped to address these dynamics, as they focus on personal data rather than inferential and predictive architectures. Another cross-cutting issue is the political economy of AI, particularly the concentration of infrastructural and epistemic power in a small number of actors. Governance discussions must therefore engage with questions of sovereignty, dependency, and digital colonialism. Finally, there is a need to integrate feminist and intersectional approaches into AI governance, not as add-ons but as structuring perspectives capable of revealing how AI systems reproduce and amplify existing inequalities.

In the European and Spanish context, AI governance is advancing rapidly at the regulatory level, yet this does not necessarily translate into effective or coherent governance. A growing gap persists between norm production and the capacity to address how AI systems restructure power, agency, and rights in practice. One critical blind spot concerns what may be described as cognitive dimensions of governance. While current frameworks emphasise safety and risk classification, they insufficiently address how AI systems shape attention, behaviour, and decision-making processes. This raises emerging concerns around cognitive liberty, particularly in environments increasingly mediated by algorithmic influence. At the same time, AI deployment in security, defence, and border management illustrates a troubling imbalance. The expansion of data-driven surveillance, predictive analytics, and automated decision-making in migration contexts has led to forms of algorithmic border governance that often operate with limited transparency and weak safeguards. These practices affect both non-European populations and EU citizens, raising serious questions regarding consent, proportionality, and the effective protection of fundamental rights. Similarly, ongoing developments around digital identity infrastructures in the European Union present both opportunities and risks. While they may enhance administrative efficiency and access to services, without robust governance frameworks they could enable forms of pervasive identification and tracking that are difficult to reverse once institutionalised. The central challenge, therefore, is not only to regulate AI systems, but to ensure that governance frameworks are capable of addressing their systemic and anticipatory implications, particularly where fundamental rights may be incrementally eroded. This also creates an opportunity for Europe to consolidate a model of AI governance that is not only compliant, but genuinely rights-constitutive, integrating emerging concerns such as cognitive liberty into its normative core.

The AI Dialogue can play a crucial role as a normative coordination platform, bridging fragmented regulatory approaches while fostering inclusive and pluralistic governance. Rather than aiming for immediate harmonisation, the Dialogue should facilitate structured convergence around core principles, particularly those grounded in human rights. This includes creating shared understandings of emerging risks, such as cognitive manipulation and systemic bias. Importantly, the Dialogue can function as a space for epistemic mediation, where different knowledge systems (technical, legal, social, and experiential) are brought into dialogue. This is essential to counter the dominance of technocratic perspectives and to ensure that governance reflects diverse realities. The Dialogue could also support capacity-building and knowledge transfer, particularly for countries with limited regulatory or technical resources. However, this must be done in a way that avoids reproducing dependency or imposing external models. Finally, it can contribute to the development of soft law instruments and best practices that, over time, may crystallise into more formal regulatory frameworks.

The AI Dialogue should build upon existing global and regional frameworks, including the UNESCO Recommendation on the Ethics of AI, the OECD AI Principles, and regulatory instruments such as the EU AI Act. These provide an essential normative baseline. However, its added value lies not in duplicating these efforts, but in connecting and critically interrogating them across scales and epistemic contexts. In this regard, particular attention should be given not only to institutional frameworks, but also to emerging research and advocacy initiatives that are already shaping the field from the margins. Projects such as DAIR's Data Workers' Inquiry, which foregrounds the lived experiences and agency of data workers, Gender Shades, which exposed structural biases in commercial AI systems, and initiatives such as Algorace, which critically examine the racialisation embedded in algorithmic systems, demonstrate that some of the most consequential insights into AI governance are being produced outside formal regulatory arenas. These initiatives do not merely complement existing frameworks; they challenge their underlying assumptions, revealing blind spots related to labour conditions, structural discrimination, and the socio-political embedding of AI systems. Bringing these perspectives into the Dialogue would allow to: 1) identify systemic inconsistencies and implementation gaps; 2) advance a more substantive form of normative interoperability; 3) integrate critical, situated, and often underrepresented forms of knowledge. Moreover, it would enable the Dialogue to foreground issues that remain marginal in dominant governance approaches, such as cognitive governance, the infrastructural conditions of AI production, and the democratic implications of AI-mediated environments. In this sense, the Dialogue has the potential to function as a meta-governance layer, not only enhancing coherence across frameworks, but also expanding what counts as relevant knowledge in AI governance.

Effective participation in the AI Dialogue requires moving beyond formal multi-stakeholder inclusion toward a structured and functionally differentiated model of engagement. Different actors should contribute according to their institutional capacities and epistemic roles. States can provide normative direction and ensure alignment with international human rights frameworks; the private sector can offer technical expertise and operational insight; academia can contribute critical and interdisciplinary analysis; and civil society can bring situated knowledge, particularly regarding the lived impacts of AI systems. However, participation must be embedded within a clear procedural architecture that mitigates asymmetries of power and influence. This could include thematic working groups with balanced representation, transparent agenda-setting processes, and iterative consultation mechanisms that allow inputs to meaningfully shape outcomes rather than merely inform them. In addition, the Dialogue should incorporate mechanisms of accountability and traceability, ensuring that contributions are reflected in outputs and that decision-making processes remain open to scrutiny. Crucially, participation should not be limited to reactive consultation but enable co-productive forms of governance, where diverse actors contribute to the definition of problems as well as solutions. This is particularly important in the context of AI, where the framing of issues (risk, safety, innovation) already carries significant normative implications. A well-designed structure would therefore allow the AI Dialogue to function not only as a forum for exchange, but as an institutional space for coordinated and reflexive governance.

Global discussions on AI governance continue to exhibit significant epistemic asymmetries, with certain voices systematically underrepresented not only in terms of presence, but in their capacity to shape agendas and normative frameworks. Among these are actors from the Global South, as well as communities directly affected by AI systems, including data workers, migrants, and populations subject to intensified forms of surveillance and algorithmic decision-making. Their exclusion reflects broader structural inequalities in access to resources, infrastructure, and institutional visibility. In addition, there is a persistent marginalisation of critical and interdisciplinary perspectives, including feminist, decolonial, and socio-technical approaches, which are essential to understanding how AI systems reproduce and amplify existing power relations. Importantly, some of the most innovative work in this space is already being produced through initiatives that operate outside traditional governance arenas, foregrounding lived experience, structural bias, and the material conditions underpinning AI systems. However, these forms of knowledge are rarely integrated into formal decision-making processes. Addressing this requires moving beyond inclusion as representation toward inclusion as epistemic integration. This entails providing material support (funding, translation, capacity-building), recognising diverse forms of expertise, and creating institutional pathways through which these perspectives can actively shape both problem definition and policy design. Without such measures, AI governance risks reinforcing the very inequalities it seeks to address.

Fostering meaningful engagement in the AI Dialogue requires complementing traditional diplomatic formats with deliberative and experimental methodologies capable of capturing the complexity and long-term implications of AI systems. One promising approach involves the use of scenario-based and anticipatory exercises, which enable participants to explore the societal impacts of AI across different temporal and geopolitical contexts. These formats can help shift discussions from reactive regulation toward forward-looking governance. In parallel, deliberative assemblies and participatory forums, including citizen assemblies or stakeholder panels, can provide structured spaces for inclusive and reasoned debate. When properly designed, such mechanisms can surface perspectives that are often excluded from formal negotiations, particularly those related to everyday experiences and diffuse forms of harm. Interdisciplinary "policy labs" or co-creation workshops may also facilitate interaction between technical, legal, and social forms of expertise, helping to bridge epistemic divides and generate more context-sensitive policy insights. Digital platforms can further expand participation, but their use should be carefully designed to ensure accessibility, inclusivity, and meaningful interaction, rather than passive consultation. Ultimately, innovative engagement should not be understood as an add-on, but as a means to enable reflexive and adaptive governance, capable of responding to the evolving and systemic nature of AI.

Effective AI governance is emerging through a combination of regulatory frameworks, institutional practices, and experimental approaches, each addressing different dimensions of the challenge. At the regulatory level, the EU AI Act represents a significant step toward a risk-based and rights-oriented framework, particularly in its attempt to classify and constrain high-risk applications. Complementary instruments, such as algorithmic impact assessments, are also gaining traction as tools to anticipate and mitigate harms before deployment. However, formal regulation alone is insufficient. Promising practices are also developing in the form of independent auditing mechanisms and oversight bodies, which seek to operationalise accountability in complex socio-technical systems and to render visible forms of harm that are often diffuse or difficult to measure. In parallel, a growing number of participatory and multi-stakeholder approaches are experimenting with more inclusive governance models. Deliberative processes, interdisciplinary policy labs, and collaborative standard-setting initiatives can enhance legitimacy and contextual sensitivity, particularly when they integrate diverse forms of expertise. Importantly, some of the most innovative contributions are emerging from critical and practice-based initiatives that foreground issues often overlooked in formal governance, such as labour conditions, structural bias, and the socio-political embedding of AI systems. These approaches highlight the need to expand governance beyond systems themselves to include the infrastructures and conditions of their production and use, as well as their effects on human agency, perception, and decision-making. Taken together, these examples suggest that effective AI governance requires not a single model, but a layered and adaptive approach, capable of integrating legal, technical, social, and epistemic dimensions while remaining attentive to emerging risks that challenge existing rights frameworks.