University of Bedfordshire

For the first Global Dialogue on AI Governance in July 2026 to be considered a success, it must move beyond high-level "principles" and deliver actionable, interoperable mechanisms. In my view, success would be defined by three key outcomes: 1. A Formal "Interoperability Framework" The most urgent risk is a fragmented "splinternet" of AI regulations. Success looks like a common set of safety standards and certificates recognized across borders. This would allow a startup in a developing nation to meet global safety requirements without navigating 193 different legal systems, effectively lowering the barrier to entry for global innovation. 2. Concrete "Capacity Commons" for the Global South Governance is hollow without access. A successful Dialogue must move from talking about the "AI divide" to solving it. This means establishing a Global Compute and Data Fund—a mechanism where advanced nations and private firms provide subsidized high-performance computing (HPC) and "sovereign data" support to the Global South. Success is seeing a researcher in Lagos or La Paz training a model on local hardware rather than just "consuming" Western AI. 3. Universal Standards for "Agentic AI" Accountability By July 2026, AI "agents" that act autonomously will be common. Success would be a global agreement on Human-in-the-Loop (HITL) baselines: a legal requirement that for every autonomous AI action—be it financial, medical, or judicial—there is a traceable human entity held accountable under international law.

• AI capacity-building
• Open-source software, open data and open AI models
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Transparency, accountability, and human oversight

My selection focuses on a "bottom-up" empowerment strategy that shifts AI governance from centralized control to local agency. The priority is to transition from passive consumption of AI to active, localized ownership. AI Capacity-building is the primary driver. True digital equity requires more than hardware; it requires a massive scale-up in literacy. By focusing on webinars, local university partnerships, and community groups, we can demystify AI and provide the technical skills necessary for independent deployment. Open-source software, open data, and open AI models are the essential tools for this mission. Proprietary, subscription-based models create financial barriers and "digital lock-in." Open-source Small Language Models (SLMs) allow individuals in the Global South to run high-performance AI on existing personal devices without recurring costs or internet dependency. Social, economic, ethical, cultural, linguistic, and technical implications are prioritized to ensure AI serves as a tool for cultural preservation. By advocating for models trained in native languages, we prevent "linguistic erasure" and ensure that AI systems respect local nuances rather than imposing a monolithic, English-centric worldview. Finally, Transparency, accountability, and human oversight ensure that as AI moves to local devices, it remains under the direct control of the user. This "on-device" approach naturally enhances privacy and ensures that humans remain the ultimate decision-makers in their own socio-economic contexts. In summary, these priorities aim to democratize AI by making it affordable, accessible, and adaptable to the unique needs of every community.

While the listed themes cover the foundational pillars of AI governance, three critical, cross-cutting issues remain under-addressed: 1. Environmental Sustainability and Resource Resilience AI's physical footprint-massive water consumption for cooling and the energy demands of hardware-is often missing from governance frameworks. For developing nations, which are frequently the most vulnerable to climate change, "AI capacity-building" must be inextricably linked to green computing. We need global standards for the energy efficiency of models and hardware "lifecycle management" to prevent electronic waste in the Global South. 2. The Transition to Agentic AI Current themes largely treat AI as a passive tool (generative AI). However, we are rapidly moving toward "Agentic AI"-systems that can independently execute tasks, move funds, and interface with critical infrastructure. This creates a "responsibility gap." A successful dialogue must address the legal personhood and liability of autonomous agents before they are deployed at scale in global financial and legal systems. 3. Epistemic Integrity and the "Reality Gap" The socioeconomic implications theme touches on this, but we need a dedicated focus on Epistemic Integrity. In a world flooded with AI-generated content, the ability of citizens to distinguish truth from synthetic fabrication is a matter of national security and social cohesion. This is not just about "transparency," but about creating a global infrastructure for content provenance and protecting the human "right to the truth." Conclusion Addressing these issues ensures the Dialogue is "future-proof." By linking AI governance to climate goals, agentic accountability, and information integrity, the UN can ensure that AI development does not inadvertently degrade our physical environment or our shared reality.

In my sector—which focuses on localized digital empowerment—the primary governance gap is the lack of a formal framework for decentralized AI capacity. Current global developments are heavily skewed toward centralized, cloud-based models, creating two significant challenges and one transformative opportunity. The Challenges "Compute Dependency" and Economic Drain: The current gap in interoperable governance means that local innovators are forced to use foreign-hosted AI services. This results in significant capital flight, as subscription fees flow out of the country, and creates a "digital leash" where local services can be throttled or disconnected based on foreign policy or price hikes. Linguistic and Cultural Erasure: Without active engagement in linguistic AI implications, my region faces the risk of "data colonialism." Most mainstream models are "fine-tuned" on Western values, leading to a loss of nuance in native language processing and traditional knowledge systems, which are not currently protected by global data standards. The Opportunities The most significant opportunity lies in the advancement of Open-Source Small Language Models (SLMs). The rapid progress in "distilled" models—which can now perform complex reasoning on standard laptops—offers a path to sovereign AI. By filling the governance gap with structured literacy programs and local institutional partnerships, we can bypass the need for massive data centers. This allows our schools, colleges, and local businesses to deploy AI that is: Privacy-First: Data never leaves the device. Cost-Free: No recurring subscription fees. Culturally Aligned: Trained and fine-tuned on local data. Conclusion: The shift toward lightweight, offline AI represents a "leapfrog" moment for developing regions, provided governance frameworks prioritize open-source protections and grassroots capacity-building over centralized cloud dominance.

The UN Global Dialogue on AI Governance aims to shift AI policy from being driven by a few powerful nations to a more inclusive global effort. It plays an important role in improving international cooperation in three key ways. First, it helps address the growing fragmentation in AI regulations across countries. At the moment, different nations are creating their own rules, which can lead to conflicts and barriers. The Dialogue provides a shared space to align basic safety standards and ethical boundaries, allowing AI innovation to move more smoothly across borders while still protecting people. Second, it gives developing countries a stronger voice. Most AI policies are currently shaped by countries in the Global North, but the Dialogue ensures that others are included in decision-making. This helps create fairer policies that consider different economic and social contexts, rather than applying the same approach everywhere. Third, it encourages practical cooperation through shared resources. This includes partnerships such as shared computing power, open-source tools, and joint research between institutions worldwide. In this way, governance becomes something that supports development, not just something that restricts it. In the end, the Dialogue's biggest contribution is building trust. By using the UN as a neutral platform, it reduces competition and helps countries work together in managing AI responsibly.

The AI Dialogue should act as a "connective tissue" that links existing but fragmented global efforts into a more coherent system. It can build on platforms like the International Telecommunication Union "AI for Good" initiative, which focuses on technical innovation. The Dialogue can help translate these projects into national policies and internationally aligned standards. It should also draw on the UNESCO Recommendation on the Ethics of AI, the most globally inclusive ethical framework, by turning its principles into practical and interoperable governance approaches. In addition, resources from the OECD AI Policy Observatory and the Global Partnership on AI provide strong research and policy insights. The Dialogue can extend this work to ensure it is relevant and accessible to the Global South. It can also bridge regional frameworks such as the European Union AI Act and emerging strategies from regional bodies, reducing regulatory fragmentation across borders. The added value of the AI Dialogue lies in its universal legitimacy. As a UN-led platform, it includes all member states, unlike smaller forums such as the G7. This allows it to support multilateral accountability, where countries align on shared safety and human rights standards. It also ensures inclusivity by addressing local needs such as language diversity and access to computing resources. Ultimately, the Dialogue can provide a pathway toward a future global AI agreement, turning today's fragmented efforts into a coordinated and trusted governance system.

To remain inclusive and effective, the AI Dialogue must move beyond a government-only approach and operate as a true multistakeholder ecosystem. Different stakeholders can contribute in distinct but complementary ways. Academia and research institutions can provide independent, evidence-based insights, including risk assessments and red-teaming results, to support scientific grounding. The private sector, including leading developers such as OpenAI and DeepMind, can contribute technical expertise, share best practices, and support capacity-building through initiatives like open-weight models and mentorship in developing regions. Civil society should act as a safeguard for public interest, ensuring that human rights, cultural diversity, and ethical considerations are embedded into global standards. Local governments can add practical value by sharing real-world use cases of AI in public services, helping bridge policy and implementation. In terms of structure, the Dialogue should adopt a hybrid and continuous model. Instead of relying only on annual meetings, it should establish year-round thematic hubs focused on key areas such as safety, linguistic diversity, and compute access. These hubs would allow technical experts and policymakers to collaborate more effectively. Regional consultation layers are also essential. Pre-summits across different regions can ensure that perspectives from the Global South are reflected early in the process, rather than added later. In addition, a formal multistakeholder advisory board with rotating membership can ensure balanced representation and faster policy response to technological change. Finally, open and continuous input mechanisms should be maintained to allow contributions as AI evolves. This approach ensures the Dialogue remains dynamic, inclusive, and closely aligned with real-world developments.

Global AI governance is still dominated by a small group of governments and large tech firms from the Global North, which leaves specific groups consistently underrepresented. First, low-resource language communities are missing from dataset design and policy discussions. This leads directly to biased systems and loss of cultural knowledge. They should be included through funded data partnerships and mandatory multilingual consultation processes. Second, Indigenous communities are excluded from decisions about how their data is collected and used. This is a governance gap, not just a participation issue. A formal Indigenous Data Sovereignty working group under the United Nations should be created with decision-making authority, not just advisory input. Third, grassroots developers and SMEs in emerging economies are absent from policy forums because of cost and access barriers. Yet they are the ones deploying AI locally. They should be included through funded regional hubs and direct representation in technical working groups, not filtered through large corporations. Fourth, Global South governments and institutions are often present but lack negotiating power due to limited technical capacity. Targeted support from bodies like the World Bank should focus on building regulatory and technical expertise so participation is meaningful, not symbolic. Fifth, youth and early-career researchers are missing from decision-making, despite long-term impact. Reserved seats on scientific panels and governance boards should be mandatory. Finally, non-technical disciplines (law, sociology, ethics) remain underweighted. Governance structures should require interdisciplinary representation rather than treating it as optional. Inclusion will not happen by invitation alone. It requires funding, formal representation, and structural changes that give these groups real influence over outcomes.

To move beyond staged diplomatic speeches, the AI Dialogue should adopt interactive, tech-enabled formats that reflect the speed and complexity of AI itself. First, policy hackathons can replace passive discussions. Mixed teams of developers, policymakers, and ethicists work together to design practical governance tools, such as prototypes for compute-sharing systems or content provenance tracking. This ensures policy ideas are technically grounded. Second, AI-enabled citizen assemblies can bring large-scale public input into the process. Platforms like Pol.is can aggregate views from thousands of participants globally, identifying areas of consensus and allowing policymakers to engage with real public sentiment rather than abstract assumptions. Third, red-team roundtables should be used to stress-test proposed regulations. One group drafts a policy, while another challenges it by identifying risks, loopholes, and unintended consequences, particularly for vulnerable communities. This improves robustness before adoption. Fourth, immersive future-state simulations can help decision-makers understand long-term implications. Using scenario mapping or virtual environments, participants can explore how policies might play out in areas such as autonomous systems or critical infrastructure. Fifth, open-mic lightning rounds can give space to grassroots innovators, especially from the Global South. Short, focused presentations allow real-world challenges and solutions to surface, breaking the hierarchy of traditional forums. Together, these formats shift the Dialogue from passive listening to active co-creation, ensuring outcomes are practical, inclusive, and responsive to real-world needs.

Effective AI governance must move from abstract ethics to functional infrastructure. Without this shift, governance risks remaining aspirational while real power stays concentrated in a few actors. First, the Public AI compute model addresses structural inequality in access to infrastructure. While frameworks like the European Union AI Act set regulatory boundaries, they do not address who can actually build AI. National initiatives such as IndiaAI and the U.S. NAIRR demonstrate how subsidised GPU access can support researchers and startups. This should be scaled into a coordinated global compute pool, potentially supported by the United Nations, enabling researchers in compute poor regions to develop locally relevant systems without prohibitive costs. Second, localised sandboxes for Small Language Models SLMs bridge policy and real world deployment. While the UNESCO Recommendation on the Ethics of AI provides guiding principles, implementation requires testing in context. Social sandboxes should enable community led pilots where SLMs run on local devices in areas such as agriculture or healthcare, particularly in native languages, allowing bias, accuracy, and cultural relevance to be validated before scale. Third, open source certification and provenance standards are critical for trust. Platforms like Hugging Face have become de facto governance hubs through tools such as model cards and datasheets, which act like nutrition labels for AI. These should be mandated for public sector use. In parallel, standards such as Coalition for Content Provenance and Authenticity provide technical mechanisms to watermark AI generated content, protecting the epistemic integrity of the information ecosystem. Together, these approaches shift governance from policing AI to enabling it, making it more equitable, practical, and globally inclusive.