CCIA

A successful first Global Dialogue on AI Governance would demonstrate a credible and inclusive multistakeholder forum that reflects how AI is developed and deployed in practice. This includes meaningful participation by industry across the AI stack and other stakeholders, on equal footing with governments, ensuring that discussions are informed by real-world development and deployment experience. Success would also be demonstrated by the Dialogue complementing, rather than duplicating, existing international efforts. It should facilitate the sharing of best practices to reduce fragmentation impeding cross-border AI development and adoption. In a rapidly evolving technological landscape, the Dialogue should discuss how domestic risk-based AI governance, rooted in international, consensus-based, voluntary AI standards, can help enable interoperability, with positive impacts for all businesses (particularly SMEs) when it comes to reducing the cost of AI adoption and use. Avoiding overly prescriptive regulatory approaches can create unintended barriers, which will be important if countries are to realize the benefits of AI. In terms of output, Chair's summaries could include curated compilations of best practices and use cases that demonstrate successful AI adoption and could follow up and assess progress with AI adoption in the lead-up to the next AI Dialogue in New York in 2027.

• Safe, secure and trustworthy AI
• AI capacity-building
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Interoperability of governance approaches

The Dialogue should cluster the themes into three buckets: AI Safety ((a) safe, secure and trustworthy AI systems, (e) protection of human rights, and (f) transparency, accountability, and human oversight); AI Opportunities ((b) capacity-building in developing countries and (c) social, ethical, cultural, and linguistic implications; and (g) open-source software, open data, and open AI models); and AI Governance ((d) interoperability and compatibility of AI governance approaches). Within those thematic buckets, priority issues are as follows: Interoperability of governance approaches is critical. Divergent national frameworks increase costs, particularly for SMEs, and create barriers to scaling AI across markets. Greater alignment can reduce fragmentation and support cross-border deployment. The social, economic, ethical, cultural, linguistic, and technical implications of AI. AI is already delivering productivity gains and enabling new economic activity across sectors. It is improving agricultural productivity through precision tools, expanding access to education through adaptive learning, and accelerating scientific discovery in areas such as health and climate, contributing to development and SDG outcomes. Industry is also expanding access to AI tools and supporting a wider range of languages and use cases, including through voluntary efforts to improve representation of underrepresented languages and contexts. Safe, secure, and trustworthy AI is foundational to the adoption and scaling of AI technologies. Industry has invested significantly in developing and implementing best practices to ensure the reliability, safety, and robustness of AI systems, recognizing that user trust is essential to broader uptake. AI capacity-building is key to uptake, particularly in developing economies, where a growing AI capacity gap risks limiting participation. Expanding access to infrastructure, skills, and data can reduce barriers to adoption. The Dialogue can build on industry efforts by identifying approaches that reduce the cost of adoption, support scalable, locally relevant deployment, and ensure governance frameworks do not constrain adoption at earlier stages of AI maturity.

One theme that is not explicitly mentioned but implicit across the seven mandated themes is the importance of AI adoption. Governments and industry that lead in AI adoption will realize the benefits, including when it comes to AI for development. With this in mind, the Dialogue should be flexible in how it engages across these themes.

Governance gaps and related developments across these thematic areas are directly affecting the ability to develop and deploy AI systems at scale, particularly across borders. A central challenge is the growing fragmentation of regulatory approaches. Divergent AI regulations increase costs for businesses seeking to develop, deploy, and scale AI products globally, particularly where national frameworks do not clearly distinguish between developers, who design AI systems, and deployers, who implement them across sectors. This lack of role clarity creates ambiguity around obligations across the AI value chain, complicating compliance and slowing adoption. Greater reliance on market-driven international standards and high-level principles can help bridge these differences and therefore reduce barriers to adoption. Other barriers arise from broader digital trade restrictions. Data localization requirements and limits on cross-border data flows constrain access to the large and diverse datasets needed to train and operate AI systems. Mandates to use local computing infrastructure can be particularly debilitating as AI-focused data centers become more specialized, needing access to cutting-edge chips that may be scarce and new electrical power and cooling requirements that may not be present in legacy data centers—construction of which involves multi-year planning. Additional barriers are emerging at other layers of the AI stack, with direct implications for the availability of representative, high-quality AI outputs. Requirements that mandate source code disclosure can deter investment. Evolving approaches to copyright and licensing are also introducing uncertainty, particularly where liability is attached to using publicly available data for training or where remuneration or licensing obligations are imposed at scale. Mandatory"pay to train" approaches applicable to publicly available data not only risk introducing systemic bias by limiting access to the data that underpins AI training and model improvement, but also are economically distortive, administratively unworkable, and likely to chill innovation without ensuring meaningful creator benefit. A better approach would target concrete harms (fraud, deception, misappropriation) while preserving and fostering legitimate learning, research, and innovation. In sum, these developments stifle innovation and access to new AI-enabled tools and services by increasing compliance burdens, particularly for firms and actors in developing economies.

The AI Dialogue should promote a shared understanding of how risk-based domestic AI regulation, grounded in international standards developed in market-driven bodies such as ISO/IEC JTC 1/SC 42, can support interoperability. This can deliver positive impacts for all businesses, particularly SMEs, and create further economic opportunity by addressing barriers to AI adoption and use, including gaps in trust and education, as well as governance fragmentation. As outlined above, divergent regulatory frameworks are already creating barriers to cross-border AI development and deployment. The Dialogue can draw on real-world implementation experience from governments, industry, and other stakeholders to inform discussions on how to mitigate the costs of regulatory fragmentation. It can support cooperation by providing a platform for ongoing engagement, evidence-sharing, and coordination across existing initiatives, without seeking to establish binding frameworks or negotiated outcomes.

The AI Dialogue should connect with existing multilateral, multistakeholder, and sectoral initiatives rather than duplicating them, focusing efforts where there are gaps in AI Governance. This includes work on AI governance within the UN as well as relevant discussions in the OECD and the G7, and regional and plurilateral efforts addressing AI governance and digital trade. The OECD's Hiroshima AI Process provides a useful, flexible reference point for advancing responsible AI in a manner that avoids more prescriptive or burdensome reporting requirements. Many of these international processes have already produced principles, frameworks, and guidelines that provide a strong foundation for cooperation on AI. The Dialogue should also connect with existing multistakeholder platforms and convenings that focus on real-world implementation, such as the Internet Governance Forum and the ITU AI for Good Summit.

Different stakeholders contribute distinct and complementary expertise, and meaningful participation by industry is essential to the effectiveness of the AI Dialogue. The private sector is responsible for developing, deploying, and scaling AI systems, and therefore brings practical insight into how these technologies function in real-world contexts. This experience is critical to ensure that discussions are grounded in AI development, deployment, and implementation, and that policy approaches are workable in practice. The Dialogue should therefore be structured as a genuinely multistakeholder process, with industry, governments, academia, and civil society participating on equal footing. This includes opportunities for all stakeholders to contribute to agenda-setting, submit written inputs, and engage in both plenary and smaller, solution-oriented discussions. Formats should prioritize interactive exchanges and allow for engagement between Dialogue sessions.

From an industry perspective, SMEs remain underrepresented in global discussions on AI governance, despite standing to gain significantly from AI adoption. SMEs rely on AI tools to improve productivity, reduce costs, and reach global markets, but are disproportionately affected by regulatory fragmentation. Barriers such as restrictions on cross-border data flows, localization requirements, and limits on access to compute can be especially difficult for smaller firms to navigate, constraining their ability to scale. Including SME perspectives would help ground the Dialogue in the realities of adoption, particularly for actors with limited resources. This can be supported through targeted outreach, dedicated consultation channels, and the use of SME case studies. At the same time, broader industry participation is necessary to reflect the full ecosystem. Larger firms develop and provide the infrastructure and services that SMEs depend on, and their participation helps ensure that governance approaches support scalability.

Meaningful engagement will depend on formats that prioritize interaction. Sessions should move beyond set-piece interventions toward solution-oriented discussions, including small group or workshop-style formats that focus on real-world use cases and implementation challenges. The Dialogue should incorporate structured opportunities for stakeholders to submit case studies, best practices, and evidence ahead of sessions, which can then inform more targeted discussions. Finally, engagement should be continuous rather than limited to single events, with mechanisms for iterative feedback and participation across Dialogue cycles, drawing from the approach of the Internet Governance Forum.

Effective AI governance is supported by policy approaches that enable innovation and deployment while addressing risks in a targeted and proportionate manner. Such policies can be enabled through forums like the Dialogue, G7, OECD, and other relevant forums that identify regulatory barriers to AI deployment, as well as lessons learned in the form of proven regulatory enablers of AI uptake. First, mechanisms that facilitate cross-border data flows are foundational. AI systems depend on access to large and diverse datasets, and policies that allow data to move across jurisdictions support both model development and deployment. Similarly, frameworks that limit requirements to localize computing infrastructure help ensure access to globally distributed compute resources. Second, protections for source code are also critical, and avoiding forced disclosure requirements helps safeguard sensitive information, support cybersecurity, and encourage continued investment in AI development. Third, in terms of broad-based AI laws, risk-proportionate, focusing oversight and due diligence on higher-risk applications rather than imposing blanket requirements on all AI systems, and building on existing sectoral frameworks as a practical path forward, with financial services and the automotive sector as examples of highly regulated sectors already deploying AI using established risk management frameworks. Finally, access to publicly available information is essential for developing accurate and reliable systems, and balanced approaches in this area can support both innovation and responsible AI.