Focor

A successful first Global Dialogue on AI Governance would establish a shared commitment to keeping humans meaningfully in control of decisions that affect their lives — not merely as a procedural safeguard, but as a foundational principle of AI governance. The most important outcome would be consensus that AI must augment human thinking, not replace it. As AI systems become more capable, there is a real risk that human judgment, creativity, and agency are progressively sidelined — not through any deliberate decision, but through the slow accumulation of automation and deference to machine outputs. Reversing or preventing this drift requires deliberate governance action. Concretely, success would mean: agreement on principles that preserve human cognitive agency; frameworks that require meaningful human involvement in consequential decisions; and commitments to monitor and address the erosion of human roles in critical sectors such as healthcare, justice, education, and public administration. Beyond this, a successful Dialogue would produce actionable recommendations — not just declarations — that member states and stakeholders can implement, along with a clear mandate for the 2027 session to review progress.

• Safe, secure and trustworthy AI
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Protection and promotion of human rights
• Transparency, accountability, and human oversight

These four themes are directly connected by a single underlying concern: the risk that AI systems erode the conditions under which humans can exercise genuine agency. Safe, secure and trustworthy AI is the baseline. Without it, no meaningful governance is possible - systems that are opaque, unreliable, or deliberately deceptive undermine every other effort. The social, economic, ethical, cultural, linguistic and technical implications of AI is perhaps the most critical theme for ensuring AI does not replace human thinking. Automation of cognitive tasks at scale is already reshaping labor, decision-making, and cultural production. Governance must grapple with the cumulative effect of these changes on human relevance and dignity - not just their individual impacts. Protection and promotion of human rights grounds AI governance in existing legal and ethical frameworks. It ensures that efficiency gains from AI cannot come at the cost of fundamental freedoms, and provides a normative anchor for contested decisions. Transparency, accountability, and human oversight is the operational mechanism for all of the above. Humans can only remain in control of AI systems if they can understand what those systems are doing and why, and if there are clear lines of responsibility when things go wrong. This theme is not just about technical explainability - it is about preserving the conditions for meaningful human judgment.

Yes. The most significant cross-cutting issue not adequately captured by the listed themes is the gradual displacement of human cognition - what might be called cognitive dependency or the atrophying of human judgment at scale. As AI systems become the default tool for analysis, writing, decision support, and problem-solving, individuals and institutions increasingly defer to machine outputs rather than developing or applying their own reasoning. This is not a safety failure in the traditional sense - the systems may be working exactly as intended - but the long-term societal effect is a reduction in human cognitive agency and, ultimately, human irreplaceability. This issue cuts across all themes: it is a social implication, a human rights concern, a transparency challenge, and a safety risk. Yet it is rarely named explicitly in governance frameworks, which tend to focus on acute harms rather than gradual erosion. A related emerging issue is AI's effect on epistemic diversity - the risk that convergence on a small number of AI systems homogenizes how people think, what information they encounter, and what conclusions they reach. This has implications for democracy, cultural identity, and scientific progress that go beyond any single thematic area. The Dialogue should consider explicitly addressing both cognitive dependency and epistemic homogenization as governance priorities in their own right.

From the perspective of a private sector technology company operating across Western Europe, governance gaps in the four selected thematic areas create both acute risks and missed opportunities. The most pressing challenge is the absence of enforceable standards for human oversight in AI-assisted decision-making. In practice, organizations - including in regulated sectors - are deploying AI systems that influence consequential outcomes without adequate requirements for human review, auditability, or accountability. The EU AI Act begins to address this, but implementation is uneven and many high-impact use cases fall into grey zones. Outside the EU, frameworks are even less developed, creating regulatory fragmentation that private sector actors must navigate at significant cost. On human rights, the gap is most visible in the lack of mechanisms to address cumulative harm. Individual AI decisions may not violate any single right, but the aggregate effect - on employment, access to services, freedom of expression - is significant and largely ungoverned. In terms of opportunities, the current window before AI governance consolidates globally represents a chance to embed strong human-centricity principles into emerging standards. Companies that build transparency and human oversight into their products by design will be better positioned as regulation tightens. There is also a real opportunity for cross-regional cooperation - particularly between the EU and the US, which has a mature tech sector and a growing role in AI development - to align on shared principles before divergence becomes entrenched. The private sector cannot fill these governance gaps alone. What is needed is clear, consistent international guidance that allows companies to invest with confidence in responsible AI practices, rather than waiting for the regulatory floor to be defined.

The AI Dialogue can play a unique role that no other existing forum currently fills: serving as a genuinely inclusive, multilateral space where governments, the private sector, civil society, and technical communities negotiate shared principles rather than simply exchanging positions. Most existing AI governance initiatives are either regional (the EU AI Act, the US Executive Order framework) or voluntary and non-binding (the G7 Hiroshima Process, the OECD AI Principles). The result is a patchwork of frameworks that creates compliance complexity for international actors and leaves significant gaps, particularly for countries outside the major economic blocs. The AI Dialogue can add value by moving beyond principles toward operational convergence. This means facilitating agreement on common definitions, baseline human oversight requirements, and mutual recognition mechanisms that allow different regulatory regimes to interoperate without requiring full harmonization. Crucially, the Dialogue should also serve as a counterweight to the concentration of AI governance influence in the hands of a small number of powerful states and companies. By providing structured participation for a broader range of stakeholders, including small and medium-sized enterprises, civil society organizations, and voices from underrepresented regions, it can ensure that governance frameworks reflect a wider set of values and interests. Finally, the Dialogue has a role in building trust. Many of the most important AI governance challenges, from cross-border data flows to the accountability of AI systems operating across jurisdictions, require states and organizations to act on the basis of shared expectations. The Dialogue is an opportunity to develop those expectations through sustained, transparent engagement rather than crisis response.

Several existing initiatives provide a strong foundation for the AI Dialogue to build upon. The OECD AI Principles and the accompanying AI Policy Observatory represent the most developed multilateral framework currently in place, with broad buy-in across member and partner countries. The Dialogue should align with and reinforce these principles rather than create a competing set. The G7 Hiroshima AI Process produced practical guidance on interoperability and risk-based approaches that the Dialogue can reference and extend to a broader membership. Similarly, the Global Partnership on AI (GPAI) has developed technical and policy expertise, particularly on responsible AI and data governance, that should be integrated rather than duplicated. At the technical level, standards work by ISO (notably ISO 42001 on AI management systems) and IEEE provides implementation-ready frameworks that the Dialogue can reference to bridge the gap between high-level principles and operational practice. The added value the AI Dialogue can bring is twofold. First, universality: unlike the OECD or G7 processes, a UN-anchored dialogue can include the full membership of the international community, giving it legitimacy that regional or club-based initiatives lack. Second, multistakeholder depth: by structuring genuine participation from the private sector, civil society, and technical communities alongside governments, the Dialogue can produce outcomes that are both politically legitimate and technically credible. The Dialogue should avoid the trap of producing yet another set of non-binding principles. Its distinct contribution should be to connect existing frameworks, identify gaps, and facilitate concrete agreements on the specific issues where fragmentation is most costly.

Effective multistakeholder participation requires more than open registration. It requires deliberate structural design that gives different groups genuine influence over outcomes, not just the opportunity to be heard. For governments, the Dialogue should maintain formal deliberative sessions with clear decision-making procedures. But the structure should also create dedicated tracks for non-governmental stakeholders, with outputs that feed directly into the main deliberations rather than running in parallel without connection. The private sector can contribute most effectively through structured technical input: sharing real-world implementation experience, identifying where governance frameworks create unworkable requirements, and piloting proposed standards. Small and medium-sized enterprises should have a distinct channel from large technology companies, whose interests and capacities are very different. Civil society organizations are best positioned to represent the interests of affected communities and to hold other stakeholders accountable. They should have a formal role in reviewing draft outputs before adoption, not only in providing initial inputs. The technical community, including standards bodies, researchers, and open-source contributors, should be integrated into working groups where technical precision matters most, such as definitions, audit requirements, and interoperability standards. An intersessional work program between Geneva 2026 and New York 2027 would allow these contributions to develop into substantive proposals rather than position statements. The Co-Chairs should publish a structured synthesis of inputs received, so stakeholders can see how their contributions were reflected and where disagreements remain.

Several communities remain significantly underrepresented in global AI governance discussions, and their absence weakens the quality and legitimacy of the outcomes. The Global South is the most significant gap. Countries in Africa, Latin America, South and Southeast Asia, and the Pacific are already affected by AI systems built elsewhere, yet have limited capacity to shape the governance frameworks that will regulate those systems. Inclusion requires more than open invitation: it requires funded participation, pre-session capacity building, and genuine weight given to their inputs in final documents. Workers and labor representatives are largely absent from AI governance spaces, despite being among those most directly affected by automation and AI-driven management systems. Their perspective on what human oversight and human dignity in AI-affected workplaces actually means in practice is essential. Older adults and people with disabilities are rarely consulted, despite being disproportionately affected by algorithmic decision-making in healthcare, social services, and public administration. Linguistically diverse communities are underrepresented partly because governance processes default to a small number of working languages. Making substantive documents available in a wider range of languages, and funding simultaneous interpretation for regional consultations, would meaningfully broaden participation. Finally, younger generations who will live longest with the consequences of current governance decisions should have structured representation, not only through youth delegate programs but through substantive roles in working groups. Building intergenerational perspective into the Dialogue's architecture would improve both its legitimacy and the durability of its outcomes.

Several existing approaches offer useful models for the AI Dialogue to draw from. The EU AI Act's risk-based framework is the most comprehensive attempt to date to create legally binding AI governance that scales requirements to the level of harm potential. While its geographic scope is limited, its methodology, tiered obligations based on risk category, mandatory human oversight for high-risk systems, and prohibited applications, provides a replicable structure that could inform international baseline standards. ISO 42001, the international standard for AI management systems, offers a practical governance tool that organizations of any size can implement. Unlike regulatory frameworks, it is jurisdiction-neutral and can serve as a common reference point across different national regimes. Encouraging its adoption globally, particularly in public sector procurement, would raise the baseline of responsible AI practice without requiring legislative action in every country. The OECD AI Policy Observatory serves as a valuable knowledge-sharing platform, aggregating national policies, incidents, and best practices in a publicly accessible format. Expanding its scope and strengthening its connection to the Dialogue's work would accelerate learning across governments. At the organizational level, the practice of publishing model cards and system cards, structured disclosures about AI system capabilities, limitations, and intended use, is gaining traction among leading AI developers. Encouraging this as a standard expectation rather than a voluntary gesture would improve transparency significantly. Finally, Focor's own work on AI-assisted knowledge management has reinforced a practical lesson: governance tools are only effective if they fit into how people actually work. The most successful approaches combine clear principles with lightweight, usable implementation frameworks. The Dialogue should prioritize practicability alongside ambition.