World Economic Forum

Moving from statements to structured follow-through The Dialogue would benefit from moving beyond declaration-driven formats such as joint statements, voluntary commitments, or principles-based declarations toward governance pathways with concrete accountability mechanisms: agreed indicators of progress, designated review bodies, and transparent reporting on implementation between convenings. Voluntary commitments can play a useful role in norm-setting, but they are insufficient without comparable evidence of implementation. The Dialogue should encourage a progression from commitments to claims to evidence: from public principles, to standardized reporting, to verifiable assurance mechanisms such as audits, certifications, sandbox exit reports, incident disclosures and independent evaluations. Ensuring that governance and regulation mechanisms account for the actual compliance capabilities of relevant actors requires sustained public-private collaboration. The Dialogue should therefore encourage structured engagement between governments and industry not only on the content of commitments, but on the feasibility of implementation, so that accountability frameworks are both ambitious and actionable. Establishing a structured link between the 2026 Dialogue and the 2027 follow-up, including agreed milestones and review processes, would help convert shared commitments into meaningful action. Centering Global Majority perspectives as agenda-shapers A successful Dialogue would ensure that the full diversity of perspectives -from emerging and developing economies as well as established technology powers- actively shapes the governance agenda rather than simply endorsing it. This matters not only for reasons of inclusion, but because effective AI governance requires integrating the various dynamics of the AI ecosystem: the interdependencies between infrastructure, data, models, energy systems, and regulatory capacity that differ significantly across contexts. Governance frameworks developed without accounting for these interconnected capabilities risk producing principles that are coherent in theory but unworkable in practice for most economies. Regional workshops conducted by the Forum and consultations conducted across Africa, Latin America, ASEAN, and India consistently underscored this gap. Connecting and rationalizing the existing governance landscape The Dialogue has an opportunity to map and connect existing AI governance initiatives rather than adding to an already fragmented landscape. Reducing duplication, clarifying relationships among national, regional, and international frameworks, and supporting interoperability across governance approaches would represent a concrete contribution. Recent analyses of responsible AI governance frameworks suggest that fragmentation, rather than the absence of principles, is increasingly the central challenge, reinforcing the need for coordination and alignment (World Economic Forum, Advancing Responsible AI Innovation: A Playbook, 2025).

• Safe, secure and trustworthy AI
• Interoperability of governance approaches
• Transparency, accountability, and human oversight

Safe, secure and trustworthy AI Safety and trustworthiness are foundational conditions for AI systems to be deployed at scale without undermining public trust or causing systemic harm. A shared, credible safety baseline- one that specifies what organizations must demonstrate about how their systems behave, fail, and are overseen- is a prerequisite for governance that is protective rather than merely declaratory. At the global level, there is not yet a common understanding of how to identify and assess emerging risks from increasingly autonomous and agentic AI systems, nor agreed methodologies for evaluating their potential impacts across different deployment contexts. Agentic AI requires dedicated governance attention because risks arise not only from model outputs, but from delegated authority, tool use, memory, multi-step planning and interaction among agents. Governance frameworks should therefore address bounded autonomy, permissioning, human-readable logs, real-time monitoring, escalation thresholds, kill-switches, post-deployment review and accountability for actions taken through agentic systems. The Dialogue could help to strengthen international alignment on risk identification and evaluation, including testing protocols, incident taxonomies and interoperable audit trails. Equally important, governance frameworks should not be structured solely around risk mitigation. Creating explicit space for articulating the benefits of agentic AI, and ensuring that frameworks actively enable safe innovation, is as important as establishing appropriate safeguards. Interoperability of governance approaches Regulatory fragmentation increasingly limits the ability of organizations to scale responsible AI innovation across borders. As divergent national and regional governance regimes multiply, locally developed AI solutions face barriers to international deployment, and the absence of a common, principle-based framework at the global level leaves actors navigating an uneven and unpredictable landscape. This fragmentation makes it difficult for innovation developed in one jurisdiction to scale globally, underscoring the continued need for a common, principle-based framework at the international level. Such an approach can provide coherence while remaining adaptable across jurisdictions. Interoperability efforts should therefore focus on building bridges across existing frameworks-through shared terminology, mutual recognition, and alignment mechanisms-rather than contributing to further proliferation, for example through crosswalks between major AI governance frameworks, common terminology for AI risks and actors, comparable assurance documentation, shared incident-reporting taxonomies, and mutual recognition pathways for testing, audit or certification where appropriate. The aim should not be identical regulation, but reduced friction between different governance systems. The Hiroshima AI Process, with which the Forum collaborates, is designed precisely as an interoperability instrument, offering a flexible, voluntary, and geopolitically neutral reference point that complements rather than competes with existing frameworks. This aligns with broader findings in interoperability-focused analyses, which emphasize the importance of shared terminology, mutual recognition, and alignment mechanisms across governance regimes Transparency, human oversight, and accountability Transparency and accountability are not peripheral governance requirements but foundational conditions for trust in AI systems. Evidence across organizations suggests that governance gaps in this area are structural: fewer than 1% of organizations have fully operationalized responsible AI practices, with ill-defined accountability structures, limited visibility into enterprise-wide AI usage, and unclear responsibility allocation across the AI value chain identified as persistent barriers (World Economic Forum, Advancing Responsible AI Innovation: A Playbook, 2025). Ef

AI sovereignty and structural economic dependencies AI sovereignty has emerged as a defining concept in national AI strategies, but its interpretation varies significantly across economies and is increasingly shaping regulatory choices in ways that may fragment the global governance landscape. Sovereignty goals are driving decisions about data localization requirements, safety and certification standards for AI models, procurement rules that favour domestic providers, and restrictions on cross-border data flows -each of which has direct implications for interoperability, market access, and the ability to build shared governance frameworks. The AI Dialogue has an opportunity to address this tension directly: building shared understanding of how sovereignty goals can be pursued in ways that remain compatible with global interoperability and the development of competitive AI ecosystems. Access to connectivity, compute, high-quality and locally relevant data, technical and governance skills, procurement pathways, financing, and the conditions for meaningful participation in model development are increasingly determinative of a country's ability to shape its technological trajectory. Open-source models and open datasets can support inclusion, but they are not sufficient on their own. They must be accompanied by documentation, safety evaluation, cybersecurity safeguards, multilingual and local-context benchmarks, and mechanisms for communities to shape deployment priorities. Efforts to expand access to compute should be paired with sustainability and transparency safeguards. Capacity-building should include not only access to high-performance computing, but also energy-efficient infrastructure, measurement of environmental impacts, responsible procurement and support for shared or public-interest compute models that reduce duplication and cost. Recent analysis, including the January 2026 World Economic Forum white paper Rethinking AI Sovereignty: Pathways to Competitiveness through Strategic Investments, highlights the need to reframe sovereignty as strategic interdependence, combining targeted domestic investment with trusted international collaboration rather than treating it as a rationale for fragmentation. Cross-cutting tensions across priorities The four priorities selected are mutually reinforcing but not without tension. Interoperability objectives- which favour common standards and cross-border openness- can conflict with sovereignty-driven regulatory choices that prioritize domestic control over data, infrastructure, or model development. Similarly, safety and transparency requirements designed for frontier AI systems may be difficult to implement in contexts with limited regulatory capacity, creating a risk that governance frameworks inadvertently exclude the economies they most need to reach. Acknowledging these trade-offs explicitly, and designing governance mechanisms flexible enough to accommodate them, will be essential to producing outcomes that are both coherent and inclusive.

Fragmentation as a systemic barrier The proliferation of AI governance frameworks at national, regional, and international levels creates compliance complexity for all organizations operating across jurisdictions - from large multinationals navigating divergent regulatory requirements to smaller enterprises and public institutions in markets where governance capacity is still developing. The result is a landscape in which the costs of compliance multiply while coherence diminishes. The implementation gap AI is driving new waves of innovation and growth. The key to realizing its full potential lies in the widespread adoption of responsible AI, the practice of building and managing AI systems to maximize benefits while minimizing risks to people, society and the environment. Yet, according to a Playbook published by the World Economic Forum in 2025, less than 1% of organizations have fully operationalized responsible AI in a comprehensive and anticipatory manner. Thus, persistent gap exists between the adoption of AI governance principles and their operational application. Awareness of frameworks does not automatically translate into implementation readiness. Such a gap presents a defining opportunity to build public trust, develop resilient markets and safeguard rights while progressing AI innovations. Evidence from international consultations indicates that stakeholders broadly welcome high-level principles but consistently call for clearer operational guidance, sector-specific examples, capacity-building support, and practical toolkits before meaningful adoption at scale can occur. The governance challenge is no longer whether AI principles are broadly recognized, but whether institutions have the tools, skills, data and accountability mechanisms to operationalize them at scale. This points to a structural shift needed in global AI governance: from framework proliferation toward implementation support. A central part of this governance gap is regulatory capacity. Many jurisdictions are not only deciding what AI rules should say, but also which institutions should enforce them, how sectoral regulators should coordinate, what technical expertise is needed, and how public authorities can evaluate AI systems in practice. The Dialogue could help countries compare institutional models, including distributed sectoral oversight, central AI coordination bodies, AI safety institutes, regulatory cooperation forums and sandbox-based learning mechanisms.

The AI Dialogue can play a distinctive role in advancing international cooperation by operating within an UN-led intergovernmental process complemented by multistakeholder engagement. It provides a dedicated venue for the full range of economies- including those that have been less represented in global AI discussions- to actively shape governance norms rather than receive them. Ensuring that priorities related to data governance, economic development, and cultural and linguistic diversity are reflected in global AI discussions will be critical to producing frameworks with genuine global legitimacy. Beyond representation, the Dialogue can serve a coordination function that existing UN mechanisms have not yet fully assumed: not simply cataloguing frameworks, but actively supporting alignment on how shared principles are implemented in practice. This means facilitating convergence on best practices for operationalizing key governance concepts- such as risk assessment, human oversight, and transparency -and supporting the capacity-building needed for those practices to be adopted across different institutional and regulatory contexts. Rather than adding new principles to an already crowded landscape, its value lies in making existing commitments more coherent, comparable, and actionable. Finally, as previously mentioned, the Dialogue can support accountability by establishing structured follow-up mechanisms -including agreed milestones, shared progress indicators, and periodic review processes - that hold participating governments accountable for translating governance commitments into domestic policy and institutional action.

World Economic Forum Centre for AI Excellence and AI Global Alliance The World Economic Forum's Centre for AI Excellence advances the responsible use of AI, data, and digital technologies to benefit society and drive industry transformation. Its work is structured around three core pillars: accelerating impactful AI innovation and adoption, preparing countries and societies for the Intelligent Era, and advancing trustworthy technology and effective governance. Together, these efforts help scale real-world AI solutions, build inclusive digital economies, and promote safe, transparent, and accountable use of emerging technologies. The Centre's AI Global Alliance brings together over 550 members across 500 organizations spanning government, industry, civil society, and academia. This initiative is advancing the safe, equitable, and responsible development of AI, fostering collaboration across sectors and borders to address governance challenges, shape international norms, and scale real-world solutions that align AI innovation with societal needs. "Resilient AI Governance and Regulation", part of the AI Global Alliance (AIGA), fosters public-private and international cooperation to shape an anticipatory, inclusive, and interoperable global AI governance ecosystem. The community of over 120 experts is currently exploring AI Agents Governance. The ""AI Competitiveness through Regional Collaboration" also part of AIGA to assist nations in cultivating competitive, resilient, and independent AI ecosystems. The initiative reinforces global dialogue among technology providers, industry players, policymakers, investors, academia, and ICT actors, leveraging the Forum's impartial platform to align diverse priorities. It is underpinned by two flagship publications: the Blueprint for Intelligent Economies: AI Competitiveness through Regional Collaboration (Davos 2025) and Rethinking AI Sovereignty: Pathways to Competitiveness through Strategic Investments (Davos 2026). An emerging model closely related to the sovereignty and infrastructure agenda is the development of digital embassies- secure digital spaces allowing nations to extend sovereign digital infrastructure beyond their borders while maintaining control over data, compute and governance. However, establishing such arrangements requires political alignment, legal clarity and robust technical and operational safeguards. The Digital Embassy Framework, spearheaded by the World Economic Forum and co-developed with stakeholders, will provide a shared baseline for designing, governing and operating trusted digital embassies worldwide The AI Dialogue may find in this emerging framework a concrete governance tool for reconciling data sovereignty with the need for cross-border cooperation. In addition, the Centre also has the Global Coalition for Digital Safety aims to accelerate public-private cooperation to tackle harmful content and interactions online and will serve to exchange best practices for online safety governance, take coordinated action to reduce the risk of online harms, and drive forward collaboration on programs to enhance digital media literacy. The work of Coalition demonstrates how collaboration between governments and industry can support the development of shared approaches to addressing AI-enabled harms across jurisdictions.

Experience from multistakeholder governance initiatives suggests several design principles that may be relevant to the AI Dialogue. Meaningful participation requires deliberate structural design. Governance processes that integrate diverse stakeholder perspectives from the agenda-setting phase tend to produce more legitimate and implementable outcomes. Regional and sectoral differentiation is important for effectiveness. A single global format may not equally serve stakeholders across different institutional contexts and levels of regulatory maturity. Complementing plenary discussions with regional and sector-specific workstreams could improve relevance and impact. Industry actors bring both implementation expertise and accountability obligations. Structured engagement around specific governance questions-such as risk management, reporting, and interoperability-can support more concrete contributions.

Several perspectives remain underrepresented in global AI governance discussions. Three gaps stand out in particular. The Global Majority: Countries across Africa, Latin America, and Southeast Asia are increasingly subject to AI systems they had no hand in governing, yet remain marginal in the forums where norms are set. Their priorities, around data sovereignty, economic development, and cultural context, rarely shape emerging frameworks, which risk defaulting to the interests of the most technologically dominant economies. Small and medium enterprises represent the majority of businesses affected by AI deployment but lack the resources to engage sustained multilateral processes. Civil society organizations face similar barriers. Addressing these gaps requires dedicated participation mechanisms, capacity-building support, and accessible formats, ensuring engagement is substantive rather than symbolic.

Two approaches may be particularly effective: Scenario-based deliberation Presenting concrete, plausible AI deployment scenarios and working backwards to governance requirements can ground abstract principles in real decisions. This approach has been effective in diverse stakeholder settings for helping participants identify shared concerns and divergences that are not visible at the level of principle formulation. Cross-regional peer learning exchanges Pairing governance practitioners from different regulatory contexts and levels of institutional maturity and regulatory contexts to share implementation challenges, rather than framework aspirations, has produced some of the most productive exchanges in regional workshop series. Participants consistently report that learning from peers navigating similar structural constraints is more actionable than exposure to governance models from jurisdictions with substantially different institutional capacities.

Multistakeholder governance models demonstrate how capacity for responsible AI can be scaled beyond individual jurisdictions by enabling peer learning, shared tool development, and collective problem-solving. Similarly, collaborative approaches between governments and technology platforms have shown potential in developing and operationalizing shared standards to address AI-related risks, particularly in areas such as online safety and content governance. Advancing Responsible AI Innovation: A Playbook The Advancing Responsible AI Innovation: A Playbook offers nine actionable, scalable and adaptable plays for turning responsible AI principles into operationalized practice. Addressing both internal organizational barriers and external ecosystem challenges such as regulatory fragmentation, it offers a concrete example of how governance guidance can be made practical and adaptive across different organizational and regulatory contexts. The Forum's AI Competitiveness through Regional Collaboration initiative, described above, offers two further governance examples. The Blueprint for Intelligent Economies demonstrates how a common framework with tailored national pathways-rather than universal prescriptions-can support countries at any level of AI maturity, including through the Regional AI Activation Networks that localize global objectives across the Middle East, Africa, and Southeast Asia. Rethinking AI Sovereignty illustrates how reframing a contested political concept-sovereignty as strategic interdependence rather than self-sufficiency can translate geopolitical complexity into actionable investment and governance choices for policymakers.