Norm Partners

A successful first Global Dialogue on AI Governance would achieve several interconnected outcomes. Agreement on core principles for responsible AI development, including transparency, accountability, and human oversight, that transcends geopolitical divides and reflects genuine multilateral consensus rather than the priorities of a few dominant actors. Meaningful engagement from the Global South, civil society, Indigenous communities, and smaller nations, ensuring that governance frameworks address diverse needs and contexts, not just those of technologically advanced states. Concrete, time-bound pledges from governments, private sector actors, and international organizations, moving beyond declarations toward measurable steps on issues like algorithmic auditing, data governance, and AI risk assessment. A clear roadmap for AI capacity-building that enables developing nations to participate in, benefit from, and shape AI development rather than simply being subject to systems built elsewhere. Incorporation of accountability and interoperability principles into existing AI regulatory frameworks such as the EU AI Act, ensuring that national and regional instruments align with globally agreed standards rather than creating fragmented compliance burdens. Establishment of a global framework on AI ethics that sets universal baseline standards for fairness, human dignity, and non-discrimination, while respecting cultural and contextual differences across nations. Creation of international AI regulatory sandboxes that allow governments, researchers, and the private sector to collaboratively test emerging technologies in controlled environments, enabling evidence-based policymaking before full-scale deployment. Including protocols for information-sharing among states when potentially dangerous capabilities emerge. Success would ultimately be measured not by joint communication but by whether the outcomes demonstrably shift how AI is governed in practice, reducing harm, expanding access, and ensuring that AI development serves humanity broadly and equitably. Establishment of shared early-warning and coordination mechanisms for frontier AI risks,

• Safe, secure and trustworthy AI
• Interoperability of governance approaches
• Transparency, accountability, and human oversight
• Open-source software, open data and open AI models

The four priorities selected reflect the foundational pillars necessary for a coherent and effective global AI governance architecture. Safe, secure and trustworthy AI is the cornerstone of any meaningful governance effort. Without baseline safety standards and trust mechanisms, AI deployment risks causing irreversible harm to individuals, societies, and critical infrastructure. Governance must ensure that systems are robust, auditable, and aligned with human values before and during deployment. Interoperability of governance approaches is essential to avoid a fragmented global landscape where conflicting regulatory regimes create compliance gaps and allow harmful AI systems to operate in jurisdictions with weaker protections. Aligning existing frameworks such as the EU AI Act with internationally agreed standards ensures coherence, reduces duplication, and enables cross-border accountability. Central to this interoperability is the harmonization of data protection and privacy standards, ensuring that personal data used to train and operate AI systems is governed by consistent, rights-based principles across all jurisdictions. Transparency, accountability, and human oversight are non-negotiable in ensuring that AI systems remain under meaningful human control. Decision-making processes must be explainable, and clear lines of responsibility must exist when AI causes harm. Strong data protection frameworks are integral to this pillar, as individuals must have the right to know how their data is collected, processed, and used by AI systems, with enforceable remedies when those rights are violated. Open-source software, open data and open AI models are critical for democratizing access to AI and preventing monopolization of transformative technologies by a small number of actors. Openness fosters innovation, enables independent auditing, and allows developing nations and civil society to participate meaningfully in shaping AI rather than merely consuming it, provided that open data practices are governed by robust privacy safeguards that protect individuals from exploitation. Together, these priorities create a governance ecosystem that is simultaneously protective, inclusive, and forward-looking, ensuring that AI development serves the collective interest of humanity rather than reinforcing existing inequalities or concentrating power in the hands of a few.

Several cross-cutting and emerging issues warrant urgent attention beyond the listed thematic areas. AI-enabled regulatory sandboxes represent an underexplored governance tool. Drawing from existing sandbox environments in financial technology and cybersecurity regulation, a comparable framework for AI would allow governments, researchers, and private actors to test high-risk AI applications in controlled settings before full deployment. Critically, these sandboxes should incorporate automated flagging mechanisms that detect and alert authorities to potentially illegal or harmful uses of AI in real time, functioning as an early warning system embedded within the governance architecture itself. Security frameworks for AI must also evolve to address the convergence of AI with other emerging technologies. The intersection of AI and blockchain, for instance, creates both new governance opportunities and new risks. Blockchain can enhance AI accountability through immutable audit trails and decentralised verification of model behaviour, but this convergence also introduces novel attack surfaces and raises questions about jurisdictional oversight that existing frameworks do not adequately address. The convergence of AI with biotechnology, quantum computing, and autonomous systems similarly demands anticipatory governance mechanisms that are not reactive but forward-looking, designed to identify and respond to risks before they materialise at scale. Additionally, the environmental impact of AI, particularly the energy and water consumption of large-scale model training and inference, remains largely absent from governance discussions despite its significant implications for climate commitments and resource equity. Finally, the governance of AI-generated content, including synthetic media and autonomous agents operating across digital platforms, requires dedicated cross-border frameworks that address misinformation, identity fraud, and the erosion of epistemic trust in ways that current thematic areas do not fully capture. Addressing these gaps would significantly strengthen the comprehensiveness and future-readiness of the global AI governance framework.

Across the countries and regions where we operate, governance gaps in the selected thematic areas are creating both significant challenges and missed opportunities for responsible AI adoption. The most persistent challenge is the absence of practical testing environments. We consistently recommend to governments and institutions the establishment of regulatory sandbox frameworks that allow AI tools and digital infrastructures to be trialled under real conditions before full deployment. This approach enables policymakers to assess outcomes, identify risks, and adapt legal and regulatory frameworks based on evidence drawn from local needs and contexts rather than importing frameworks designed for entirely different environments. Our portfolio companies already offer operational solutions across many of the thematic areas under discussion, including tools addressing safe and trustworthy AI deployment, transparency mechanisms, interoperability infrastructure, and data governance. Safeguarding these innovations within a coherent and predictable regulatory environment is essential to ensuring they can scale and deliver impact across diverse jurisdictions. The rapid advancement of agentic AI systems, capable of autonomously executing complex multi-step tasks across industries, is creating profound and largely ungoverned disruptions to labour markets. Unlike previous waves of automation, agentic AI displaces not only routine tasks but increasingly cognitive and professional functions, disproportionately affecting mid-skilled workers in developing economies where social protection systems are weakest. Governance frameworks must urgently address workforce transition, reskilling obligations, and the equitable distribution of productivity gains generated by autonomous AI agents. AI-driven media narrative control represents another critical gap. The use of AI to manufacture and target disinformation at scale poses direct threats to democratic institutions and social cohesion, requiring mandatory disclosure requirements and independent oversight mechanisms. Finally, the integration of AI into defence technology is outpacing existing international humanitarian law, demanding urgent multilateral dialogue to establish red lines and accountability standards before deployment becomes irreversible.

The AI Dialogue can serve as a uniquely legitimate and neutral multilateral platform capable of bridging the fragmented landscape of AI governance initiatives that currently operate in silos across regions, sectors, and institutional mandates. Its most critical role is to function as a convergence mechanism, bringing together outputs from existing national, regional, and multi-stakeholder processes into a coherent global framework that no single body has yet been able to achieve. Rather than creating parallel structures, the Dialogue should synthesise existing commitments and identify binding or voluntary measures that command genuine cross-regional support. The Dialogue can also advance international cooperation by establishing shared technical standards and mutual recognition agreements that allow AI systems developed in one jurisdiction to be assessed and deployed in others without duplicative compliance burdens. This is particularly important for developing nations that lack the regulatory capacity to independently evaluate complex AI systems. Critically, the Dialogue must go beyond norm-setting and create operational cooperation mechanisms, including joint incident reporting systems, coordinated responses to AI-enabled threats such as disinformation and autonomous weapons, and shared infrastructure for AI safety research. International cooperation at this level requires trust-building measures that only a UN-convened process can facilitate.

The AI Dialogue should build upon several existing initiatives while adding distinct value that current mechanisms cannot provide. The OECD AI Principles and the Global Partnership on AI provide foundational normative frameworks and technical working groups that the Dialogue should formally incorporate rather than duplicate. The EU AI Act represents the most advanced binding regulatory instrument and offers a legislative template that the Dialogue can adapt into interoperable international standards. UNESCO's Recommendation on the Ethics of AI provides a globally endorsed ethical baseline spanning 193 member states and should serve as the ethical foundation of the Dialogue's outputs. The Hiroshima AI Process and bilateral AI safety agreements between major powers offer precedents for state-level coordination that the Dialogue can multilateralise. The African Union's Continental AI Strategy and similar regional frameworks from ASEAN and Latin America must be actively integrated to ensure the Dialogue reflects governance priorities beyond the Global North. The added value the AI Dialogue uniquely brings is legitimacy at universal scale. No existing initiative combines the participation, mandate, and authority of a UN General Assembly process. This positions the Dialogue to do what others cannot: convert fragmented commitments into a coherent global architecture, establish an inclusive sandbox cooperation network for cross-border AI testing, create binding accountability mechanisms, and ensure that the voices of the most affected communities, including those in developing economies facing acute labour market disruption from agentic AI, are central to shaping the rules that will govern their futures.

The AI Dialogue should adopt a multi-stakeholder architecture that moves beyond traditional intergovernmental formats to reflect the reality that AI governance requires input from those who build, regulate, deploy, and are affected by AI systems. Governments should participate as primary norm-setting actors, but the private sector, civil society, academia, and technical communities must have structured and substantive roles, not merely observer status. The format should include dedicated thematic working groups that report directly to plenary sessions, ensuring technical expertise informs high-level political commitments. Roundtable sessions should be central to the Dialogue's format, deliberately designed to maximise perspective density by bringing together voices that rarely occupy the same room: frontier AI researchers, startup founders, regulators, civil society leaders, labour representatives, and policymakers from the Global South. The richness of governance outcomes is directly proportional to the diversity of perspectives stress-testing proposed frameworks in real time. Startups and emerging technology ecosystems must be recognised as essential participants rather than peripheral observers. They are frequently the first to identify governance gaps, develop practical solutions, and operate at the intersection of innovation and risk. Dedicated startup and ecosystem tracks within the Dialogue would ensure their insights directly inform norm-setting rather than arriving too late to shape outcomes. Bridge builders, individuals and organisations with credibility across both government and the AI and emerging technology sectors, are indispensable to the Dialogue's success. These actors translate between the language of policy and the realities of technical development, preventing governance frameworks from becoming disconnected from the systems they seek to regulate. Their formal inclusion as facilitation actors, not just participants, should be a structural feature of the Dialogue. Regional preparatory dialogues should precede the global convening, and a permanent digital participation infrastructure should ensure meaningful remote engagement for those unable to attend in person.

The most significantly underrepresented voices in global AI governance discussions include communities from the Global South, Indigenous peoples, persons with disabilities, workers facing displacement from agentic AI, startups and emerging technology ecosystems, and youth. Developing nations, despite being among the most profoundly affected by AI-driven economic disruption, are frequently marginalised in technical discussions dominated by major powers. Their inclusion requires dedicated capacity-building support, translation resources, and guaranteed substantive roles rather than token representation. Indigenous communities hold governance frameworks rooted in collective rights and intergenerational responsibility that offer vital correctives to predominantly individualistic AI ethics paradigms. Their inclusion requires culturally appropriate engagement processes developed in partnership with Indigenous-led organisations. Startups and innovation ecosystems, particularly those operating in emerging markets, represent a perspective that is systematically underrepresented despite being closest to the practical realities of AI deployment. Their absence from governance discussions produces frameworks that reflect the concerns of large incumbents rather than the full breadth of the sector. Workers facing labour market displacement from agentic AI systems represent a constituency whose lived experience is largely absent from governance forums. Trade unions and labour advocacy groups must have formal seats at the table. Debate must be embraced as a productive governance tool rather than a procedural inconvenience. Contested perspectives, when engaged rigorously and respectfully, produce stronger and more legitimate frameworks than consensus processes that paper over fundamental disagreements.

The AI Dialogue should move decisively beyond traditional panel discussions toward formats that generate genuine insight, build trust, and produce actionable outcomes through perspective density and structured debate. Roundtables should serve as the primary engine of the Dialogue, deliberately convened to maximise the density and diversity of perspectives in each session. Small, carefully curated groups combining startup founders, government officials, civil society representatives, technical experts, and affected community members create the conditions for the kind of frank exchange that larger plenary formats suppress. The quality of governance frameworks produced is directly linked to the quality of deliberation that precedes them. Structured debate formats should be formally incorporated, with stakeholders assigned to argue and defend positions under scrutiny rather than simply presenting prepared statements. This approach surfaces hidden assumptions, tests the resilience of proposed frameworks, and builds the mutual understanding necessary for durable international agreements. Live regulatory sandbox demonstrations would allow governments and innovators to engage with AI tools operating within governance frameworks in real time, grounding abstract policy discussions in practical reality. Bridge builders with credibility across government, the AI sector, and emerging technology ecosystems should be formally embedded as facilitators within these sessions, ensuring that technical realities and policy ambitions remain connected throughout the Dialogue. A persistent digital platform operating between formal sessions would maintain momentum, enable ongoing collaboration, and ensure that the startup and innovation ecosystem can contribute continuously rather than only during formal convenings.

Several policies, practices, and platforms offer concrete models for effective AI governance that the Dialogue should examine and build upon. The EU AI Act represents the most comprehensive binding regulatory framework currently in force, establishing a risk-tiered approach that calibrates regulatory requirements to the severity of potential harms. Its interoperability provisions and conformity assessment mechanisms offer a replicable template for jurisdictions seeking to develop their own frameworks while maintaining alignment with international standards. Regulatory sandbox programmes pioneered in financial technology regulation, notably by the UK Financial Conduct Authority and the Monetary Authority of Singapore, demonstrate how controlled testing environments can generate the evidence base needed for adaptive, outcomes-driven regulation. Translating this model into AI governance, with embedded flagging mechanisms for illegal or harmful applications, would allow regulators to stay ahead of rapidly evolving capabilities rather than perpetually catching up. The OECD AI Policy Observatory provides a valuable platform for cross-country policy comparison and knowledge sharing, demonstrating how multilateral institutions can support governance capacity-building without imposing binding obligations prematurely. Within our own portfolio and operational experience, we have observed that the most effective governance approaches share three characteristics. They are co-designed with the private sector and startup ecosystems rather than imposed upon them. They incorporate sandbox environments that allow local adaptation before national legislative embedding. And they deploy bridge builders with credibility across both government and the technology sector to ensure that policy frameworks remain technically grounded and practically implementable. On emerging risks, platforms that use AI to detect and flag AI-generated disinformation in real time, and blockchain-based audit trail systems that create immutable records of AI decision-making, represent promising technical governance solutions that the Dialogue should actively evaluate and potentially endorse as components of a global best practice framework.