European Union

A successful first Dialogue should lay the foundation for greater international cooperation on AI while balancing urgency with inclusivity, scientific rigour, political commitment and a shared understanding of key principles. It should provide a clear and operational working method to ensure the UN and its MS have the means to deliver on their vision of AI as a global public good, building on existing initiatives and on the complementary strengths of all stakeholders. Effective inclusive multistakeholder participation is essential to ensure enough opportunities for exchanges but also to translate into improved cooperation between actors. Success requires not just diverse representation but dedicated moments to foster greater visibility, exchange and cooperation, such as opportunities to present initiatives, matchmaking spaces, and others – while ensuring synergies with AI for Good Summit and WSIS Forum. Government and private sector engagement should go beyond statements. The Dialogue could be an opportunity for clarifying shared concerns for AI development/applications, and to set in motion or scale up ideas for concrete solutions that contribute to making safe, inclusive and globally accessible AI a reality, leveraging notably the potential of open source. The Dialogue should also help to put AI discussions on a credible, independent scientific baseline. One element for this will be the report by the Scientific Panel. Success here means all stakeholders using science as a common knowledge base or reference point for future discussions, avoiding fragmented approaches. The first Dialogue should create the conditions for effective collaboration over the long term by fostering mutual understanding, identifying practical areas for cooperation, and establishing durable channels for future engagement. Success will be measured by the depth of, and breadth of participation in the conversations, the clarity brought to the challenges ahead, and the demonstration of willingness of participants to continue working together in structured and transparent ways.

• Safe, secure and trustworthy AI
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Interoperability of governance approaches
• Protection and promotion of human rights

1. SST AI: to prevent harm from unreliable, malicious systems or misuse, we need transparency, accuracy, information integrity, cybersecurity, and accountability so citizens and businesses can trust and confidently adopt AI technologies. But without transparency, accountability, and human oversight, safe, secure and trustworthy AI will remain an aspiration rather than a realistic AI governance goal. 2. HR: AI systems must be designed, developed and deployed in ways that respect and promote dignity, equality, and human rights, as well as democratic principles. The EU therefore advocates for safeguards by design and a human-centric approach to AI. Online protection of women and girls, who are disproportionately affected by AI-facilitated gender-based violence, as well as the protection of children and youth against dangerous online practices, should be absolute priorities. 3. Interoperability: ensuring a certain level of compatibility of approaches, building on existing binding frameworks, would help prevent regulatory arbitrage between jurisdictions, especially with regards to systemic and universal risks, while fostering the diffusion of good practices and facilitating responsible cross-border deployment of AI. We should pave the way for a stronger articulation of the multitude of international initiatives, to better leverage their respective strengths and complementarities in a coordinated manner, based on shared principles and approaches. 4. Implications of AI: all stakeholders should exchange on the broader implications of AI on jobs, culture and creation, language diversity, environment protection, and economic fairness. The Dialogue should discuss how AI ecosystems can be inclusive from the onset. The impact of AI on the labour market is of particular interest to us. We support global linguistic and cultural heritage, and want to ensure AI-driven growth benefits all. This also raises key questions about equitable access to data and the fair distribution of value derived from linguistic and cultural resources within AI systems.

The framing of the priorities disregards the positive impact AI can have on economic development and societal good. AI is a powerful force for positive change. By fostering responsible AI innovation and an AI First Policy, the EU is unlocking transformative potential in key sectors, promoting efficiency, safety, transparency, or pollution reduction. Beyond industry, the EU is democratising AI's benefits by integrating generative AI into the public sector. AI holds significant potential in education, including in remote and underserved areas. The Dialogue should also specifically address the issue of child safety. Another priority currently missing is the interlinkage between AI and our planet. On the one hand, the carbon footprint of AI, alongside its environmental lifecycle, risks undermining global climate and sustainability goals if left unchecked. AI's energy demands are soaring and are set to rise heavily. The Dialogue should make sure to follow 'Green AI' principles, AI systems and models which have a reduced carbon, material, water and broader environmental footprint alongside their lifecycle, whilst promoting the resilient, efficient and circular use of sustainable energy, water and hardware. This should also include circular economy considerations. On the other hand, AI also offers new tools to pursue sustainable development and well-being, with significant potential to accelerate progress towards the SDGs, including efforts to combat climate change ('AI for Green'). The Dialogue can build on the recent UNEA resolution on "environmental sustainability of AI systems". Existing initiatives, such as the Coalition for Digital Environmental Sustainability (CODES) or the Coalition for Sustainable Artificial Intelligence could be encouraged to further materialise proposals in both objectives. Another subject that is inherent but not explicit is ways of AI training and diffusion within regions in order to be closer to the people and create dialogues within communities or across communities in the same region.

Bridging governance gaps is a prerequisite for global AI governance to achieve its intended objectives. The EU has substantial experience in harmonizing policy approaches across EU Member States in various digital fields and overcoming governance challenges in the area of AI. The EU is ready to share its experience with other countries through the AI Global Dialogue. In particular, the EU has a lot to share with regards to open-source approaches in AI which should be treated as an overarching strategic priority. They can enable more equitable participation in the development of meaningful AI applications as digital public goods, where stakeholders can actively engage in AI innovation and oversight. The EU stands ready to mobilize its private sector and civil society, including actors of open source and open weights AI solutions, to develop global partnerships that can contribute to enhancing access to safe and inclusive AI for the public interest.

The Dialogue can leverage shared scientific foundation, reducing fragmented approaches and aligning global discussions. This shared baseline could support the identification and prioritisation of targeted collaborative initiatives, enabling stakeholders to align efforts around concrete, evidence-based problem statements. The Dialogue can also foster trust through transparency and multistakeholder exchange. By creating structured opportunities for knowledge-sharing, the Dialogue can break down silos and information asymmetries. The Dialogue should particularly showcase solutions from developing countries so that other regions can learn from their experience. There is no "one-size-fits-all" approach but we need to disseminate cross-cutting learnings more quickly. To ensure that individuals, organisations, and institutions can engage with AI confidently, critically, and safely, AI capacity-building must be recognised as a cornerstone of effective AI governance – underpinning trustworthy AI, the protection of fundamental rights, and the interoperability of governance frameworks. Without robust human and institutional capacity, even the most well-designed policies risk remaining declarative rather than operational. A comprehensive approach is required to strengthen capacity at all levels, moving beyond basic AI literacy to foster informed agency – empowering stakeholders to evaluate, question, and deploy AI systems responsibly. Capacity-building is not merely a standalone priority but a prerequisite for achieving meaningful and sustainable outcomes in all thematic areas of AI governance. By investing in skills, systems, and mindsets, we can ensure that AI is developed and deployed in ways that are responsible, inclusive, and aligned with societal values. Finally, the Dialogue can translate political and private sector engagement into actionable outcomes. Success lies in understanding shared risks and areas of concern and identifying low-hanging fruit for immediate collaboration. Ultimately, the Dialogue's role is to create the conditions for sustained collaboration. By fostering mutual understanding, practical cooperation, and durable engagement channels, it can lay the groundwork for a more inclusive global AI governance.

• The UNESCO AI Ethics Recommendation and its annual Forum • The ITU and its annual AI for Good Conference • The ITU AI Skills Coalition • UN Interagency Woking Group on AI (as far as the UN is concerned this is an already established mechanism for coordination). • The UN Independent Scientific Panel on AI • World Summit on the Information Society (WSIS) and its annual Forums • The Internet Governance Forum (IGF), especially the Policy Network on AI (PNAI) • EU Independent Scientific Panel on AI • EU Advisor Forum • Hiroshima AI Process and Friends Group – Voluntary Code of Conduct • Global Partnership on AI (GPAI) • Artificial intelligence safety institutes • The AI Summit series (the next summit organisers) • The Group of 7 • The Group of 20 • Current AI • Children Online Protection Lab • Laboratory for Women's rights online • Coalition for Digital Environmental Sustainability • Coalition for Sustainable Artificial Intelligence • UNFCCC AI for Climate and their AI for Climate collection in partnership with DPGA • The Hamburg Sustainability Conference and the Hamburg Declaration on Responsible AI for the SDGs • Work of OHCHR on AI • AI HUB for sustainable development (launched by Italy during its G7 Presidency, and UNDP) Potential added-value of the Global Dialogue: Unlike sector-specific forums or high-level principles (e.g., UNESCO AI Ethics Recommendation, OECD Recommendation on AI), the Dialogue has the potential to bridge science, policy, and multistakeholder exchange in a single platform. It combines the rigour of expert assessments with political momentum and global inclusive participation. This Dialogue could amplify and connect existing initiatives and strive for interoperability among them, including by leveraging complementarities and helping articulate the currently fragmented international AI governance landscape.

The Global Dialogue should be inclusive, and therefore empower diverse stakeholders to contribute meaningfully, including on different panels and through opportunities for proper exchange amongst all stakeholders, on an equal footing. The EU also supports the idea of pairing one MS with one stakeholder for the co-leading of sessions, to ensure that different perspectives are reflected. Separate sessions for governments and other stakeholders do not seem useful, and multi-stakeholder attendance should be possible during all sessions. In line with Resolution 79/325, the high-level governmental segment should be part of a multistakeholder plenary session, and should be very short. The EU recommends mixing speaking slots for governments and other stakeholders in all sessions, as already practiced during the Launch Event of the Dialogue last year. All interested stakeholders should get a chance to voice their views at the Dialogue. Consequently, enough time needs to be allocated for this. It is therefore recommended to move away from plenary sessions and allow more parallel sessions, possibly also more than two sessions at the same time. Cross-cutting sharing of learnings works best when we overcome silos. One first step could be that stakeholders expose their approaches or solutions to an "interoperability check" by different stakeholders. We advocate for formats that include stakeholders from different backgrounds. This way we get direct cooperation between otherwise disconnected stakeholders. In this context, the EU can draw on its experience with large scale participatory processes that combine inclusive consultation formats and deliberative exchanges. This experience could be complemented by structured breakout discussions and clear methods for synthesizing inputs into actionable recommendations, ensuring that stakeholder contributions effectively inform the Dialogue's outcomes. The Secretariat should also ensure that access (e.g. accreditations…) is facilitated and that information is shared broadly and on time.

Voices and perspectives from developing countries could be included further. This is an important priority for the EU, in particular under the new International Digital Strategy. The EU is already cooperating on AI with developing countries partners through various channels. For example, through the D4D Hub advisory groups, we collaborate with civil society and private sector stakeholders. In addition, the EU is developing a Tech Business Offer for partner countries, combining private and public investment with policy solutions. In the LAC region, we work with partners through the EU-LAC Digital Alliance, for example on building a supercomputing network for artificial intelligence that powers innovation across Europe, Latin America and the Caribbean. The voices and perspectives of children as well as vulnerable and underserved communities must also be reflected. It is essential that intergovernmental and multistakeholder segments are equally weighted in both design and outcomes, with no implicit hierarchy between them. This balance is critical to ensure that the Dialogue reflects not only governmental perspectives, but also the expertise of all stakeholders including civil society, private sector, academia and the technical community – especially from developing countries. In its first iteration in Geneva, the Dialogue should already address how the meaningful participation of all stakeholders can be ensured for the second Dialogue which will take place on the sidelines of the STI Forum 2027 in New York.

"Audience-driven discussions" could be organised where the audience submits questions in advance through online tools such as Slido in one morning. The most popular questions form the agenda for that afternoon panel discussion or workshop. This would allow also stakeholders that could not be present physically to voice their concrete concerns while still allowing the experts to prepare detailed answers. "Matchmaking session" as a guided, short (10-15') one-on-one or small-group sessions between unlikely pairs. Participants would sign up for this session on the day, and an algorithm would assign them a pair from a different sector. They would be provided with suggested guiding questions such as "What's one AI governance challenge you think the other side underestimates, and why?". The idea is to break echo chambers and help identify blind spots. This could also encourage greater cooperation between projects and initiatives. Panel sessions should be steered by a moderator so that panelists react to each other's interventions. Panelists should not start with prepared statements, and there should be no more than 3 or 4 panelists in any panel. There should be a possibility for the audience to put questions to the panelists without prior inscription. The Partner2Connect interactive workshop at last year's AI for Good Summit may be an example to follow . We also need more testimonials and objective research on AI effects and scalability beyond simple risk analysis and mitigation. There is a common understanding that "one size does not fit all" but we lack the middle layer on which elements can be effectively transposed from one use case to others. The global dialogue could therefore include interactive formats where concrete questions are put before groups of stakeholders for them to propose spontaneous solutions. To increase innovative aspects, the participants could be assigned to groups randomly.

• The EU AI Act is the AI governance framework for the whole European market. It follows a risk-based approach, introducing rules for the development and use of AI systems that are commensurate to the risks. • The EU is also coordinating the development of Codes of Practice as voluntary guidelines to support compliance (multistakeholder policy tools). • This is complemented by strong governance at EU level: - The AI Board gathers all EU MS around the implementation of the legislation at national level, and discusses AI related challenges when they arise. - The Independent Scientific Panel to support the scientific understanding of AI capabilities and advise the EU on emerging risks. - The Advisory Forum, a general advisory body complementing the work of the scientific panel. - The AI Office has a comprehensive mission, reflecting the EU's strategy on AI. - The AI Pact is a multistakeholder platform for companies and organisations to prepare for the different stages of implementation of the AI Act. • The AI Challenges (European Institute of Innovation and technology) is a way to accelerate the deployment of cutting-edge AI solutions across industries and public sectors. • The Council of Europe Framework Convention on AI and human rights, democracy and the rule of law - first-ever international legally binding treaty in this field. • The Hamburg Declaration on Responsible AI for the SDGs, the first ever multistakeholder declaration explicitly linking AI with the 2030 Agenda and sustainable development. • The OECD AI observatory has a collection of policies, use cases and tools and metrics all related to AI governance. It could become the basis for a global repository. • OHCHR should be a standing partner of the Dialogue, and the Dialogue could serve as a harbour for OHCHR's work on AI - anchoring it in the UN HR system and linking outcomes to the HRC.