AI Governance Initiatives and Approaches: Experiences, Lessons and Complementarities

Mataboge argues that Africa must move beyond being a passive consumer of AI technologies developed elsewhere and instead become a co-creator and architect of AI systems. She emphasises that AI adoption on the continent must serve developmental goals — addressing concrete societal and economic problems — not just corporate interests where commercial incentives already drive adoption.

She noted that African leaders at the dialogue were increasingly advocating for a 'just AI transition', centred on how Africa will emerge as an architect of AI systems rather than remaining a consumer of externally developed technologies . She further stressed the need to ensure AI adoption is not only 'AI for corporate Africa' but also 'AI for developmental Africa', solving concrete problems in broader society and the economy .

The African Union unveiled its continental AI strategy in 2024 to provide a common framework so that all 55 member states do not pursue entirely separate paths to AI adoption and regulation. The uptake has been strong, with 16 African countries already having developed their own national AI strategies, demonstrating significant continental appetite for engagement with the AI agenda.

Mataboge stated that the AU continental AI strategy, unveiled in 2024, aimed to ensure member states have common guidelines rather than 55 separate paths to AI adoption and regulation . She noted that 16 African countries have already developed their own national AI strategies, reflecting strong appetite from the continent .

Rather than operating as 55 individual markets, Africa's greatest opportunity is to build a continent-wide AI ecosystem where countries collaborate and enhance each other's capabilities at scale. Regional champions — countries with foundational infrastructure such as stable and affordable energy — can anchor shared technological infrastructure that benefits not just their own country but the wider region.

Mataboge identified South Africa, Kenya, Nigeria, Egypt, Rwanda, and Uganda as examples of regional anchors for AI adoption, each with foundational infrastructure capable of supporting broader regional ecosystems . She noted that there is strong consensus that Africa's AI trajectory will be strengthened by anchoring and strengthening these regional champions .

Mataboge outlined five priority areas that Africa must address to avoid being left behind in the AI transition. These span physical and digital infrastructure, African control over data and models, representation of African languages and contexts in AI systems, long-term investment approaches, and improved coordination across policymakers, private sector, and civil society.

She enumerated the five priorities as: strengthening infrastructure and compute capacity through shared regional models; advancing sovereign capability including African control over data, infrastructure, and model development; closing gaps in data and representation particularly for African languages; aligning capital with long-term ecosystem development through patient capital and public interest investment; and improving cross-sector coordination to translate intent into execution .

Egypt has developed a risk-based governance model that concentrates regulatory attention on high-stakes areas such as human rights, children, data protection, and public trust, rather than treating all AI systems identically. This tiered approach makes governance affordable and scalable, particularly for countries with limited regulatory capacity.

Baraka described Egypt's fourth-tier model as concentrating attention where stakes are highest - rights, human rights, children, data protection, and public trust - rather than treating every system identically, making governance affordable and not just principled . She also outlined the full implementation stack: national AI strategy, institutional governance through the National Council for AI, implementation capacity through the Egyptian Centre for Responsible AI, a governance framework, operational guidelines, and AI procurement guidance .

Baraka argues that effective AI governance requires careful sequencing rather than rushing from principles to regulation. Governance must be understood as an ecosystem — encompassing strategy, institutions, standards, procurement, and real deployment — all of which must advance together rather than in isolation.

She stated that Egypt did not leap from principles to regulation but built a stack, starting with a national AI strategy, then institutional governance, then implementation capacity, then a governance framework, then operational guidelines, and finally AI procurement guidance . She concluded that AI governance is an ecosystem - not just a document, not just a strategy, not just institutions - and that strategy, institutions, standards, procurement, and real deployment must advance together .

Baraka contends that international cooperation is most effective and durable when it is built around tangible shared resources rather than shared declarations or principles. Concrete tools such as common language models, reference governance frameworks, or shared diagnostic instruments give countries ongoing reasons to remain engaged long after a meeting concludes.

She stated that 'cooperation strengthens fastest around shared resources and not shared statements', citing a common language model like CARNAC, a reference governance framework, or a shared diagnostic tool as examples that create reasons for countries to stay in contact long after meetings end . She also referenced Egypt's AI Share initiative in cooperation with UNDP as a mechanism for sharing use case applications with neighbouring countries in Africa and the Arab region .

Baraka emphasises that well-designed governance frameworks are insufficient on their own; they require the human and institutional capacity to actually implement them. This includes trained personnel, testing and audit functions, and readiness assessments for both institutions and AI systems.

She stated that 'frameworks do not implement themselves' and require trained people, testing and audit functions, and readiness assessments for both institutions and systems . She cited Egypt's Applied Innovation Centre as the implementation arm that enables use cases to be deployed in priority sectors such as health, agriculture, and education .

Masaki argues that international AI governance cooperation can be effective even when countries adopt different policy approaches, as long as they share a commitment to human-centric values. The OECD supports this by fostering common understanding of key concepts, trends, and challenges grounded in robust evidence.

He stated that 'effective international cooperation does not require identical approaches' and that what matters most is a shared commitment to human-centric values guiding the development and deployment of AI . He noted that the OECD supports this by fostering a common understanding of key concepts, trends, challenges, and opportunities grounded in robust evidence .

Masaki identifies three areas of growing convergence in the international AI governance landscape. These are: recognition of AI as a high government priority with shared objectives despite different policy approaches; broad understanding that cross-border AI systems require international cooperation and interoperability; and the increasing importance of shared values and guardrails as AI systems grow more powerful.

He described growing convergence in three dimensions: first, AI as a high priority for governments with shared objectives of spreading benefits widely and addressing risks; second, broad understanding that cross-border AI systems require international cooperation and interoperability; and third, the growing importance of shared values and guardrails as AI systems advance in capabilities such as supporting scientific discovery and finding security vulnerabilities . He also noted that AI adoption reached 52% of large firms compared with just 17% of small firms across the OECD in 2025, illustrating the uneven adoption landscape that governance must address .

The Global Partnership on AI (GPAI) functions as a bridge connecting technical expertise with policymaking, research with implementation, and regions with different capacities and priorities. It brings together a broad coalition of countries and expert communities to support evidence-based international cooperation on AI governance.

Masaki noted that the Global Partnership on AI currently brings together 46 countries informed by a multidisciplinary expert community . He described GPAI and its broad expert networks as serving as a bridge between technical expertise and policymaking, between research and implementation, and between regions with different capacities and priorities .

The OECD has developed practical tools to help governments move from principles to implementation, including the AI Policy Toolkit and the forthcoming OECD AI Index. These tools were developed through international cooperation and are designed to be flexible enough to respect national circumstances while grounding policy in evidence.

Masaki announced the launch of the AI Policy Toolkit, which helps governments identify barriers to AI adoption and translate common principles into practical, evidence-based policies that reflect national contexts . He also noted that the OECD AI Index would be launched later in the year to enable countries to measure their progress in trustworthy AI policymaking . He further referenced the revision of the Hiroshima AI Process Reporting Framework as a mechanism giving organisations a simple way to show how they are implementing trustworthy AI .

The European Union has developed a comprehensive four-dimensional AI strategy that addresses AI adoption across society and the economy, the provision of public compute infrastructure, a comprehensive legislative framework through the EU AI Act, and international cooperation. Each dimension is considered essential and complementary to the others.

Viola outlined the four dimensions: first, fostering AI adoption through an 'AI first' principle requiring all organisations to consider AI for new transformation projects ; second, creating public compute infrastructure including 19 AI factories and larger AI gigafactories across Europe ; third, the EU AI Act providing a risk-based legislative framework based on proportionality and necessity ; and fourth, international cooperation in multilateral and bilateral forms .

Viola argues that the complexity and capital intensity of AI development, combined with its profound societal implications, makes it impossible for any single jurisdiction to govern AI effectively in isolation. The EU has therefore invested significantly in international outreach alongside its domestic framework, seeking a common narrative for AI governance.

He stated that 'the Union really believes that AI governance can be effective only if it is a shared effort' . He noted that the EU invested in a comprehensive framework while also reaching out internationally to find a common narrative for AI, acknowledging that the answer is not yet completely clear and that humility and collective effort are required . He also observed that recent scrutiny of powerful AI models has led jurisdictions previously favouring private-led initiatives to consider more comprehensive frameworks, indicating growing convergence .

Mujica argues that ISO's primary contribution to AI governance is the provision of international standards that create a universal common language, agreed methodologies, and technical requirements applicable across all jurisdictions. This universality is what makes standards a uniquely powerful tool for enabling global interoperability.

He explained that standards bring a universal and common language - agreeing on definitions of terms such as AI, transparency, and risk - as well as agreed methodologies for processes such as impact assessment, and technical requirements that carry a promise of expected outcomes . He stated that the same standard can be used in Zimbabwe, Egypt, Chile, Germany, the US, and China, creating the basis for interoperability in a way that builds trust .

Mujica clarifies a common misconception by distinguishing standards from policy and regulation. Standards do not create policy or regulation but rather provide the bridge to implement those policies and regulations in practice, defining the 'how' while policymakers define the 'what'.

He stated clearly that ISO does not do policy - that is the job of policymakers - and does not create regulation, but provides a bridge to implement those policies and regulations in real-world practice . He illustrated this with the example of incident reporting: ISO would not tell anyone to whom an incident should be reported (that is the job of regulators), but can provide a framework for data governance, what information should be collected, and how to put it together - the 'how' .

Mujica argues that standards can enable genuine interoperability even when countries maintain divergent national regulations, by providing shared frameworks that can be applied universally. The example of incident reporting illustrates how standards can define the data governance and collection methodology without prescribing the regulatory destination of that data.

He stated that even with divergent policy frameworks, shared standards frameworks can provide a real plug-and-play environment . Using incident reporting as a concrete example, he explained that ISO would not prescribe to whom an incident should be reported, but could provide a framework about data governance, what information should be included, how to collect it, and how to compile it - supporting the 'how' while policymakers define the 'what' .

Mujica argues that without consistent and verifiable standards, organisations in different countries may claim to follow responsible AI practices without any means of verification, creating unfair competitive advantages. Conformity assessment frameworks address this by providing a trusted mechanism for verifying compliance.

He noted that without consistency, organisations make efforts to do things in a certain manner but there is no way to know what others are doing, which can create unfair competitive advantages . He referenced a discussion with ARSO, the standard body for the African region, to illustrate the goal of 'one standard with one test, with one certificate recognised everywhere' as what truly creates interoperability .

Mujica articulates the ultimate ambition of international standardisation as achieving a situation where a single standard, tested once and certified once, is recognised universally. This would eliminate duplication, reduce compliance burdens, and create genuine interoperability across borders.

He cited a previous exchange with ARSO, the standard body for the African region, to articulate the shared ambition: 'one standard with one test, with one certificate recognised everywhere', describing this as what really creates interoperability through standards .

Mujica challenges the assumption that a standard's work is complete upon publication, arguing that real impact is only achieved through implementation and demonstrated positive outcomes on the ground. This requires partnerships between standard-makers and policymakers rather than parallel, siloed processes.

He stated that standard-makers 'live under the illusion that our job is done when a standard is published', but in reality the job is done only when the standard is implemented and has created a positive impact on the ground . He cited the ISO policy and standards AI journey, which combined forces with Korea, the Netherlands, and many international organisations to create a space where policymakers and standard-makers work together, with Malaysia and Egypt basing their national AI strategies on this journey .

Ibáñez highlights that Latin America and the Caribbean is already seeing significant AI activity in the public sector, with over a thousand pilots underway, demonstrating real momentum and appetite for AI adoption. However, the region still lacks the foundational infrastructure — including digital connectivity, institutional capacity, and governance frameworks — needed to scale AI responsibly.

She noted that the region has more than a thousand pilots running in the public sector alone, acknowledging this is likely an underestimation . She cited fixed broadband penetration at only about half the level of OECD countries, and noted that only seven out of 26 borrowing member countries score above 50% in the AI Readiness Adoption and Governance Index . She also cited the example of Brazil's state of Ceará, where IDB support helped modernise courts with AI, lifting judicial productivity by 40% .

Ibáñez argues that Latin America and the Caribbean does not need to build AI governance institutions from the ground up, as the region already has significant institutional foundations that can be adapted to the demands of the AI era. She points to existing data protection laws and dedicated authorities as examples of foundations that can be leveraged and evolved.

She stressed that the region 'does not start from scratch' and does not need to reinvent existing institutions but rather evolve and adapt them to new technologies . She cited data protection as a case in point, noting that 17 countries in the region already have data protection laws and 12 countries have dedicated authorities .

Ibáñez argues that technological sovereignty for the LAC region does not mean self-sufficiency or doing everything independently, but rather retaining the capacity to govern data, infrastructure, and AI in the public interest through a combination of national capabilities and regional coordination. Interoperability and standards are therefore crucial.

She stated that 'for the region, technological sovereignty is not about doing everything alone' but about coordinating and retaining the capacity to govern data, infrastructure, and AI in the public interest through a combination of national capabilities and regional coordination and cooperation . She emphasised that interoperability and standards would be crucial to achieving this .

Ibáñez argues that blended finance mechanisms and public-private partnerships are essential tools for enabling countries in the LAC region to invest in the digital infrastructure needed for responsible AI adoption. These mechanisms help share risk, extend investment horizons, and direct private sector resources towards connectivity and digital infrastructure for underserved populations.

She described blended finance as a tool to help countries share risk, extend investment horizons, and reinvest private sector resources into connectivity and digital infrastructure, which is crucial for accelerating inclusion of underserved populations and low-income areas . She noted that the IDB group works with governments and the private sector to help scale and coordinate these efforts across the region .

Ibáñez argues that the IDB's unique value in supporting AI governance lies in two assets that take decades to build: granular technical knowledge of each individual economy and country, and the programmatic capacity to act across many countries and sectors simultaneously. This combination enables the IDB to act as a convener and financier across the public-private continuum.

She identified the IDB's two key assets as granular technical knowledge of each economy and country, and the programmatic capacity to act across many countries and sectors simultaneously . She noted that the IDB acts across the public-private continuum through three windows - IDB, IDB Invest, and IDB LAC - and that its whole-of-government approach paired with deep sectoral presence allows it to act at scale .

Ibáñez presents a significant economic opportunity for the LAC region from AI adoption, but cautions that realising these gains requires deliberate investment in foundational conditions. Drawing on historical analogies, she argues that transformative technologies only deliver productivity gains when institutions and ways of working are fundamentally redesigned around them.

She cited IDB estimates suggesting that AI in the LAC region could add 5% to regional output over the coming decade through the labour channel alone . She drew on the historical analogy of electricity, noting that it did not increase productivity until factories were redesigned around it, arguing that AI will be no different unless the state and private sector transform how they actually work .

Zhang argues that AI governance frameworks must be grounded in a deep understanding of how the technology is actually evolving, rather than being designed only for current capabilities. The rapid shift in industry focus — from language models to world models and agentic AI — means that governance frameworks risk becoming outdated before they are even finalised.

She noted that the narrative of AI is shifting rapidly in the industry, with discussions moving from language models and chat in the previous year to world models and agentic AI in the current year . She argued that understanding where technology is heading is critical to ensure governance frameworks are not only designed for what already exists but also prepare for the future .

Zhang contends that while much governance discussion focuses on the risks of AI models themselves, the greater and more varied risks emerge at the point of deployment, particularly in highly regulated industries. The risk profile for a specific use case in healthcare, for example, is fundamentally different from that in financial services or insurance.

She stated that 'more risk present when we deploy AI' compared to the risk profile of the model itself, and that 2025-2026 is characterised by large-scale AI deployment . She cited her experience as a board member of Common Spirit Health Foundation, one of the largest non-profit healthcare systems in the US, which deploys a hybrid mix of large language models, small language models, in-house solutions, and third-party tools, requiring a governance layer across the entire system rather than for any single model .

Zhang argues that while some elements of AI governance — such as evaluation methodologies and transparency requirements — can and should be standardised globally, deployment governance must be tailored to reflect the specific risk profiles of different industries and regions. A single uniform approach to deployment governance would be impractical and potentially counterproductive.

She stated that 'the regulation of governance cannot be identical for different industry, different region' and that making governance dynamic and diverse is very important . She elaborated that for evaluation and transparency, standardisation is appropriate, but for deployment, governance must be practical and reflect different risk profiles across different countries and industries .

Zhang argues that technology itself can be leveraged as a solution to governance challenges, rather than governance being conceived solely as a constraint on technology. Federated computing is cited as a mature example that enables AI training on sensitive data without requiring physical data transfer, thereby addressing compliance concerns in highly regulated industries.

She cited federated computing as an example of technology that has become increasingly popular in highly regulated industries like healthcare and finance, because it allows data owners to share data for AI training without physically transferring or moving their data, thereby addressing compliance concerns . She questioned whether such technological solutions are being considered in governance and regulation discussions .

Zhang proposes reframing AI governance from a compliance burden to a competitive advantage, arguing that if responsible AI is positioned as a market differentiator rather than merely a regulatory requirement, founders and deployers will have stronger intrinsic incentives to build and deploy trustworthy AI. This would create a positive feedback loop benefiting the entire ecosystem.

She questioned why governance frameworks could not create incentives for founders and innovators to build responsible AI, rather than treating it only as a regulatory requirement . She argued that if responsible AI becomes a differentiation and competitive advantage rather than just a regulatory requirement, there will be more incentive for founders to implement governance, and industry leaders deploying AI will prefer vendors that have already built responsible AI in .

Pielmeier distils three essential elements for effective multi-stakeholder AI governance from the lessons of Internet governance: inclusive processes that bring in all relevant stakeholders; common normative frameworks that provide a shared language for what is to be achieved and avoided; and clear focus on the priorities of communities most acutely impacted by AI.

He stated that effective governance requires inclusive process, common normative frameworks, and clear focus and priorities . He elaborated on each: inclusive process through forums like the Internet Governance Forum; international human rights law as the common normative framework ; and focusing on communities most impacted and facing highest risks to establish priorities .

Pielmeier argues that AI governance should draw on the extensive experience accumulated through Internet governance processes rather than starting from scratch. The Internet Governance Forum and the São Paulo Principles from the Net Mundial process represent valuable institutional knowledge and normative frameworks that can directly inform AI governance.

He noted that AI is built largely on top of the social and technical architecture of the Internet, which has undergone more than 20 years of process development, discussion, and multi-stakeholder coordination . He referenced the Net Mundial process led by the Brazilian government and the Brazilian Internet Steering Committee, which produced the São Paulo Principles as guidance for constructing multi-stakeholder processes, arguing these should guide all AI governance commissions and processes . He also invited participants to put the 2026 Internet Governance Forum in Nairobi on their calendars .

Pielmeier argues that international human rights law already provides the shared normative framework needed to guide AI governance, and that this should be explicitly recognised and built into governance processes. This includes not only core international conventions but also frameworks that translate human rights into private sector responsibilities.

He stated that international human rights law provides the normative framework for understanding what AI governance seeks to achieve and avoid, and that this should be said out loud and built into processes . He cited the UN Guiding Principles on Business and Human Rights and the OECD Guidelines on Multinational Enterprises as frameworks that translate human rights into responsibilities for the private sector, which have been embedded through processes like the Global Network Initiative into internal corporate governance .

Pielmeier argues that meaningful inclusion of civil society from the global majority goes beyond token representation at governance forums. These organisations must be genuinely involved in setting the agenda, shaping priorities, and preparing the substance of governance discussions, particularly given that they represent communities most impacted by AI.

He described the goal as ensuring civil society organisations from the global majority are 'not just a seat at the table in internet governance conversations but really are in the kitchen helping to set the menu, helping to prepare the meal, helping to serve the needs of the stakeholders that are often most excluded and overlooked when we convene in places like Geneva or New York' . He argued that focusing on communities most impacted and facing highest risks should help establish governance priorities .

Pielmeier describes a concrete initiative launched by the Global Network Initiative in partnership with the Centre for Communications Governance at the National Law University in Delhi, designed to build bridges so that civil society perspectives from the global majority can be fed into AI governance processes at the international level.

He described the initiative called 'Multi-Stakeholder Approaches to Participation in AI Governance', launched around the AI Impact Summit in India, which is intended to ensure civil society organisations from the global majority have genuine participation in AI governance conversations . He noted that the initiative brings resources and attempts to build bridges so that these perspectives can be fed into processes including those of the UN, the Swiss government, and their upcoming summit .

Pielmeier emphasises that the AI governance community should resist the temptation to create entirely new institutions and processes when existing ones — developed through decades of Internet governance work — can be adapted and built upon. New institutions should only be created when there is a specific need that cannot be addressed through existing efforts.

He stated that there are organisations and institutions that have been working on digital governance for years, including ISO, OECD, and many others, and that 'we should remember that and build on those foundations' rather than trying to create new institutions unless there is a specific need that cannot be addressed through existing efforts . He noted that AI models are built from data scraped from the Internet, deployed through the Internet, and rely on the Internet to interoperate, making Internet governance experience directly relevant .

Li describes the World Economic Forum's Centre for AI Excellence as a global platform dedicated to accelerating the responsible use of AI, data, and digital technologies. Its governance work focuses on advancing trustworthy technology through forward-looking frameworks and multi-stakeholder collaboration, positioning the WEF as an impartial convener for these discussions.

She stated that the Centre for AI Excellence serves as a global platform to accelerate the responsible use of artificial intelligence, data, and digital technologies, with its AI governance work focused on advancing trustworthy technology and effective governance, promoting safety, transparency, and accountability through forward-looking frameworks and multi-stakeholder collaboration .

Lamanauskas introduces the Enhanced UN AI Resource Hub as a comprehensive, system-wide tool that brings together AI initiatives from across the UN system into a single accessible platform. The hub is designed to serve as a one-stop shop for member states seeking capacity-building support, fellowship opportunities, and information on what the UN is doing across the AI spectrum.

He stated that the hub, unveiled in December at the General Assembly and developed by UNDP together with ITU and UNESCO, already brings together more than 1,000 AI initiatives reflecting contributions from 55 diverse UN entities . He described it as a comprehensive place to find what the UN is doing across the AI spectrum, with the latest iteration specifically including capacity-building opportunities, fellowships, and contact points .

Lamanauskas explains that the UN Interagency Working Group on AI received a new mandate through the WSIS Plus 20 outcome document to systematically map AI capacity-building initiatives across the UN system, identify gaps, and explore how existing capabilities can be leveraged. The Enhanced UN AI Resource Hub is described as one of the first practical steps in implementing this mandate.

He stated that the interagency working group received a new task in paragraph 86 of the WSIS outcome document resolution to map out AI capacity-building initiatives including fellowships, look for gaps, leverage existing capabilities across the UN system, and report back to the Global Dialogue on AI . He described the Enhanced UN AI Resource Hub as one of the first practical steps implementing this mandate .

Gabriel presents key findings from the hub's data, highlighting that public institutions are the primary beneficiaries of UN AI initiatives, while training programmes and outcome reporting remain comparatively underdeveloped relative to tools and research products. However, she notes a positive trend of growing fellowships and training opportunities compared to the previous year.

She noted that public institutions are the primary beneficiaries with 581 mentions, with scope to engage civil society, academia, the private sector, and communities more consistently . She reported that 508 AI tools and 498 research products dominate the portfolio, while training and outcome reporting remain comparatively limited with 152, though a trend of more trainings and fellowships is emerging compared to the previous year's report .

Opp articulates the core purpose of the UN AI Resource Hub as making the UN system's collective knowledge more accessible to member states, reducing duplication of effort across UN entities, and helping countries identify the most appropriate support for their specific AI governance and capacity-building needs.

He stated that the purpose of the UN AI Resource Hub is simple: 'to make the system's knowledge more accessible to avoid duplication and support member states to identify the right support for the AI governance and capacity building needs' . He noted that the hub now features a dedicated page on AI capacity-building offers and fellowships, providing a consolidated entry point for member states and UN colleagues .

Opp emphasises that effective UN support for AI governance must be tailored to the diverse circumstances of member states, recognising that countries are at very different stages of digital development and have different priorities, capacities, and contexts. Support must also be grounded in rights-based and development-oriented principles.

He stated that countries need support that is 'timely, practical, rights-based, and development-oriented' and that also 'reflects their different starting points, capacities, priorities, and contexts' . He noted that around the world, countries are moving quickly to develop AI strategies, strengthen institutions, and build regulatory capacity for applying AI responsibly .

Opp frames the hub not merely as an information repository but as a mechanism for strengthening the coherence and coordination of the UN system's collective offer to member states. The goal is to move from separate, siloed initiatives to a connected ecosystem of support that can turn collective knowledge into coordinated action.

He stated that the hub is 'a way to strengthen the collective offer of the UN system, not as separate initiatives, but as more connected ecosystems of support' . He articulated the aspiration that 'if we use it well, the UN AI Resource Hub can help turn our collective knowledge into coordinated action and ensure that AI serves people, development, and public good' .