Norwegian University of Science and Technology

(i) Concrete coordination across existing governance efforts (OECD, G7, Council of Europe, AI summits) rather than adding another layer of duplication. (ii) Agreed international standards on what digital or AI sovereignty actually means in practice. (iii) Actionable outputs beyond a communiqué: specific interoperability mechanisms, capacity-building commitments, and a clear roadmap for the Scientific Panel. (iv) Meaningful roles for civil society, technical experts, and affected communities in the main sessions, not just side events. (v) Address the risk that open-source AI, while democratising access, may still embed the norms and biases of its originating countries. (vi) Addressing the military and intelligence applications of AI raising questions about strategic autonomy and national security.

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

(i) Safe, secure and trustworthy AI is the baseline. Without safety standards, all other governance efforts are meaningless. Public trust depends on AI systems functioning reliably without causing unintended harm. (ii) Social, economic, ethical, cultural, linguistic and technical implications of AI reflect the full scope of disruption. AI is reshaping labour markets, concentrating economic power, marginalising non-dominant languages, and challenging cultural norms. Governance cannot be narrowly technical while societal consequences are ignored. (iii) The protection and promotion of human rights provide the normative anchor. International human rights law offers an existing, universally recognised framework covering privacy, non-discrimination, freedom of expression, and access to remedy. This ensures governance serves people, not just industry or state interests. (iv) Transparency, accountability, and human oversight make the other three operational. Safety commitments and human rights protections are only meaningful if there are clear obligations to explain how AI systems work, assign responsibility when harm occurs, and keep humans in the loop.

(i) Digital and AI sovereignty- No theme directly addresses the growing tension between nations' right to govern AI within their borders and the reality that AI systems, data flows, and infrastructure are controlled by a small number of foreign corporations and states. This is especially urgent for developing countries with no domestic computational capacity or foundational models. (ii) Concentration of power- The listed themes do not explicitly address the unprecedented concentration of AI development in a handful of private companies. This raises questions about democratic governance, market competition, and whether states can meaningfully regulate entities that increasingly outmatch them in resources and technical expertise. (iii) Environmental sustainability- AI's growing energy and water consumption, reliance on rare minerals, and expanding hardware footprint are absent. Governance frameworks that ignore the environmental costs of AI risk accelerating ecological harm, particularly in countries hosting data centres or mining operations that do not benefit equally from AI.

(i) Norway has strong institutions and data protection laws, but depends heavily on foreign cloud providers and AI models, creating sovereignty risks. AI systems perform poorly in Norwegian and Sami languages, threatening linguistic diversity. Rapid data centre expansion lacks environmental governance. Regulatory fragmentation across Europe creates uncertainty for cross-border AI deployment. Norway's opportunity lies in leveraging its transparent institutions and sovereign wealth to pioneer responsible AI standards and fund global capacity-building. (ii) India faces AI deployment in welfare, credit, and policing without adequate oversight or redress. Foundational AI development remains concentrated abroad despite India's large technical workforce. Labour displacement is an urgent threat given the scale of the services sector. Hundreds of Indian languages are poorly served by current AI systems. However, India's massive domestic market, growing startup ecosystem, and rising multilateral influence create significant opportunities.

It can act as a coordination hub rather than a competing initiative. The AI governance landscape is fragmented across regional blocs, bilateral agreements, and voluntary commitments. The Dialogue is well-positioned to map these efforts, identify overlaps and gaps, and promote interoperability between national and regional frameworks without imposing a single model. The Scientific Panel's annual reports can establish a shared evidence base. Much international disagreement on AI governance stems from different assessments of risk, capability, and impact. Independent, rigorous scientific assessments can ground policy discussions in common facts, reducing the space for purely political positioning. The Dialogue can amplify voices that are currently absent. Developing countries, indigenous communities, linguistic minorities, and civil society organisations rarely influence AI governance at the international level. A well-designed Dialogue process can ensure their concerns shape outcomes rather than being treated as afterthoughts. It can build practical bridges on specific issues even when a broad consensus is impossible. Geopolitical tensions between the US, China, and others will not be resolved in this forum. But targeted cooperation on AI safety testing, capacity building, data sharing for public goods, and language diversity is achievable and valuable.

The OECD AI Principles and G7 Hiroshima Process have established norms for responsible AI, but remain limited to wealthy nations. The Council of Europe AI Convention is the only binding treaty, but it applies only to a narrow group of signatories. The Dialogue can extend these frameworks to the majority of countries currently left out. The AI Safety Institute's network and AI Action Summit series have advanced technical safety work. The Dialogue can ensure safety standards reflect global concerns, not just the priorities of a few leading AI nations. UNESCO's AI ethics recommendation has broad adoption but poor implementation. Regional strategies from the African Union, ASEAN, and others remain underfunded. The Dialogue can identify implementation gaps and mobilise support where it is most needed. Civil society coalitions and the Global Partnership on AI bring research and citizen perspectives but lack institutional channels to influence policy. The Dialogue can provide that channel. The added value is simple: no other forum combines universal membership, an independent scientific panel, and a recurring structure. The Dialogue can be the connective tissue that turns fragmented initiatives into a coherent global approach.

Governments should share concrete national experiences with AI regulation, including failures, rather than repeating broad commitments. Developing countries must be supported with funding and preparatory resources so their participation is meaningful. The private sector should provide honest technical briefings on frontier capabilities and risks, disclose governance practices, and include smaller companies from diverse regions alongside dominant firms. Civil society and academia should bring independent research, impact assessments, and the perspectives of affected communities to challenge assumptions and hold powerful actors accountable. The Scientific Panel should ground every session in evidence, setting the agenda around emerging risks rather than political convenience. All of these contributions are undermined if the format relies on large plenary speeches. The Dialogue should prioritise smaller thematic working groups, require written submissions in advance, and focus discussion time on genuine disagreements and practical solutions. Annual convenings alone are too slow for a fast-moving technology, so intersessional consultations, regional preparatory meetings, and online engagement should maintain momentum between sessions. Most importantly, each session should produce specific, trackable action items with clear ownership rather than consensus declarations that avoid hard questions.

The most affected communities are often the least heard in AI governance. Workers facing automation-driven displacement, indigenous peoples whose languages and knowledge systems are ignored by AI, communities subject to algorithmic decisions in welfare and policing, women and gender-diverse groups affected by encoded biases, small enterprises from developing countries governed by rules shaped for large corporations, farmers and healthcare workers using AI tools in low-resource settings, and young people who will live longest with today's governance choices all remain underrepresented. Inclusion requires more than invitations. It means funded participation, multilingual engagement, accessible formats, preparatory support, structured testimony processes, and genuine influence over outcomes rather than decorative presence at side events.

The Dialogue should break from traditional UN formats, such as plenary speeches and panels. Structured scenario exercises where delegations work through real governance dilemmas can surface practical solutions faster than prepared statements. Live demonstrations of AI systems, including their failures, would ground abstract discussions in tangible reality. Citizen assemblies with demographically representative participants can bring public perspectives directly into the room. Red-teaming sessions, in which experts stress-test proposed frameworks, can reveal weaknesses early. Reverse briefings, where affected communities present their experiences to policymakers, can shift power dynamics and surface blind spots. Intersessional digital platforms should maintain momentum between annual meetings, so in-person time focuses on resolving real disagreements. Regional preparatory convenings can ensure positions reflect broad consultation. Whatever formats are chosen, they must genuinely influence outcomes rather than being innovative paths to the same weak declarations.

The EU AI Act is the most comprehensive risk-based AI regulatory framework globally. It categorises AI systems into unacceptable, high, limited, and minimal risk tiers, imposing corresponding obligations from outright bans to transparency requirements. It addresses foundation models by requiring dedicated provisions for general-purpose AI, including safety evaluations, documentation, and transparency for the most capable systems. The Act demonstrates that binding, cross-sector AI legislation is achievable at scale and provides a regulatory template that other jurisdictions can adapt. For the Dialogue, it offers lessons on balancing innovation with protection, on how risk classification can structure governance discussions, and on how enforcement mechanisms can be built across diverse national contexts within a common framework. Norway, as an EEA member, aligns with the EU AI Act while adding distinctive national practices. Its transparency requirements for public-sector AI allow citizens to see which automated systems are used in decisions affecting them, offering a practical accountability model that goes beyond the EU baseline. Norway's approach to AI in welfare services emphasises human oversight and the right to explanation, ensuring algorithmic decisions can be challenged. The Norwegian Data Protection Authority has taken an active role in scrutinising AI deployments, setting precedents on facial recognition and profiling. Norway also invests in language technology to ensure AI works in both Norwegian and Sami, addressing linguistic inclusion that most governance frameworks overlook.