University of Embu

In my opinion, the primary indicator of success for the first Global Dialogue would be the formalization of a Universal AI Risk Evaluation Framework. This would move beyond voluntary commitments toward a shared technical standard for assessing frontier models, ensuring that safety is not a luxury good exclusive to the Global North. Secondly, success requires the establishment of a Global AI Capacity Fund. A dialogue that ends without a concrete financial and technical roadmap for bridging the compute divide fails the Global South. Success means securing commitments for regional compute hubs and the transfer of AI expertise to developing nations, ensuring they are co-creators of AI, not just consumers. Finally, success would be the achievement of Regulatory Interoperability. With AI governance currently fragmenting into regional blocs such as EU AI Act vs. US Executive Orders a successful dialogue must yield a common denominator for transparency and human oversight. This ensures that AI developers face consistent ethical requirements globally, preventing governance arbitrage where high-risk systems are deployed in under-regulated jurisdictions.

• Safe, secure and trustworthy AI
• AI capacity-building
• Open-source software, open data and open AI models
• Social, economic, ethical, cultural, linguistic and technical implications of AI

These themes represent the four pillars of a stable AI future and require immediate intervention: Safe, Secure, and Trustworthy AI: Urgent action is needed to establish standardized red-teaming and watermarking for synthetic content. As AI-generated misinformation threatens democratic integrity globally, we must move from trust to verifiable safety through mandatory independent audits of high-risk systems. AI-Capacity Building: This is the most pressing equity issue. Engagement must focus on Knowledge Transfer 2.0, moving beyond basic literacy to high-level technical training and infrastructure grants. Without this, AI will exacerbate Sustainable Development Goals (SDG) gaps. Socio-economic, Ethical, Linguistic, and Technical Implications: Active engagement is required to prevent linguistic erasure. Most LLMs are trained on Western data; urgent action is needed to curate high-quality datasets for indigenous and under-represented languages to ensure AI remains culturally relevant and ethically grounded. Open Source, Open Data, and Open AI Models: Open-source is the ultimate check against market monopolization. Engagement here should protect the right to open innovation, ensuring that small-scale developers and researchers have the data and model access necessary to build localized, transparent solutions.

In my opinion, the current themes lack sufficient focus on agentic autonomy and environmental sustainability. Agentic AI & Chain of Accountability: We are rapidly moving from chatbots to agents that can execute financial transactions and manage infrastructure. A critical emerging issue is the accountability gap - the legal and technical challenge of assigning liability when an autonomous agent makes a catastrophic error. This requires a new category of governance centered on agentic auditing and kill-switch protocols that are not yet fully captured in current trustworthy AI discussions. Environmental & Resource Governance: The focus on capacity building often ignores the physical cost. Training frontier models consumes millions of liters of water for cooling and massive amounts of energy. An emerging issue is the ecological AI footprint. Future governance must include standards for green AI, requiring transparency in the carbon and water intensity of model training. Without this, the digital advancement of some nations may come at the direct environmental cost of others, creating a new form of climate injustice.

Impact on Kenya & East African Region In Kenya, the primary challenge is the Regulatory Lag. While we have the National AI Strategy (2025–2030) and the AI Bill (2026), our local governance is struggling to keep pace with the rapid deployment of AI in high-stakes sectors like marine conservation and blue economy management. The Biodiversity Data Paradox In the conservation sector, we are seeing a surge in nature tech using AI for real-time acoustic monitoring of coral reefs and satellite-driven anti-poaching efforts. However, a significant gap exists in data sovereignty. Much of the biodiversity data collected in the Western Indian Ocean is processed by AI models owned by Global North entities. Without robust open-data frameworks, we risk digital biopiracy, where local ecological insights are commodified without returning value to the coastal communities who protect these resources. Linguistic and Technical Exclusion A major hurdle in our region is the linguistic compute gap. Current AI models lack the granularity to process indigenous knowledge systems or local Swahili dialects effectively. This creates an emerging form of exclusion where AI-driven climate adaptation tools are black boxes that local fishers and farmers cannot trust or influence, leading to a breakdown in the social contract of technology. The Sovereign AI Opportunity Conversely, the move toward sovereign AI infrastructure evidenced by the 2026 Nairobi AI Forum presents a transformative opportunity. By investing in regional compute hubs and open-source models tailored to tropical ecosystems, East Africa can lead in green AI. We could leapfrog legacy conservation methods, using AI to manage marine protected areas with unprecedented precision, provided we close the governance gap between technological adoption and ethical oversight.

In my view, the Global Dialogue should not just be another meeting; it needs to be the connective tissue for a world where AI regulation is currently scattered. Right now, we have different regions building their own AI islands, which creates a fragmented landscape. The Dialogue's most vital role is to move us from simply talking about ethics (the "what") to building shared frameworks (the "how"). As an academic leader, I see the Dialogue as a bridge. It ensures that AI standards are not just handed down by tech-heavy nations but are co-authored by the Global South. It advances cooperation by: • Creating a Common Language: We need a universal baseline for safety so that responsible AI means the same thing in Nairobi as it does in New York. • Bringing Science to the Table: By leaning on the International Scientific Panel on AI, we ensure policy is based on technical reality, much like how we use science to drive climate action. • Championing Inclusion: This is the only space where every nation has a seat. It's where we can ensure that cooperation includes actual resource sharing such as compute power and data rather than just selling software to the rest of the world.

We do not need to reinvent the wheel. There is already brilliant work happening, and the AI Dialogue should act as a master synthesizer for these efforts. It should specifically plug into: • UNESCO's Ethics Recommendations: This is a fantastic foundation for human rights that we can now move into the "monitoring and results" phase. • The ITU's AI for Good: This is where the practical, hands-on work for the Sustainable Development Goals (SDGs) is happening. • Regional Strategies: Like the African Union's AI Strategy, which understands the specific linguistic and economic nuances of our continent. The Added Value: While groups like the G7 or the OECD do deep technical work, they often lack universal legitimacy. The real "added value" of this Dialogue is its convening power. It takes the best ideas from smaller, high-income groups and vets them through a global lens to see if they work for everyone. Most importantly, it breaks down silos. Usually, the scientists and the politicians are in different rooms. This Dialogue brings them together into one recurring, predictable process. It turns AI governance from a series of "one-off summits" into a living, breathing global conversation that evolves as fast as the technology does.

Stakeholder Contributions and Structure AI governance is too complex for any one group to manage alone. Stakeholders should contribute through a Triple-Helix model: • Academia & Civil Society: Act as the ethical compass, providing independent audits, impact assessments, and research on indigenous knowledge integration. • Private Sector: Contributes technical feasibility and red-teaming insights to ensure regulations don't stifle innovation. • Governments: Provide the democratic mandate and enforcement mechanisms. For the format, I recommend a hub-and-spoke structure. The hub is the central UN Dialogue, while the spokes are regional consultations such as in Nairobi, Bangkok, and Bogota. This ensures that local nuances such as Kenya's specific labour laws or educational gaps inform global policy. This should be a rolling dialogue, not a one-off event, with a digital platform for real-time submission of successes and failures from the field.

Underrepresented Voices and Inclusion Currently, we are seeing a Global South silence. Specifically underrepresented are: • Indigenous Communities: Whose languages and traditional knowledge are often excluded from training data. • The Human-in-the-Loop Workforce: The gig workers and data annotators (many based in Africa and SE Asia) who build AI but have no say in its governance. • Micro-SME Developers: Small-scale innovators who lack the legal teams to navigate complex regulations. How to include them: We must move beyond observer status. I propose Travel and Technical Grants to ensure these voices can physically attend sessions. Additionally, we should implement linguistic quotas for consultations, requiring that a percentage of evidence be gathered in local languages to break the English-centric nature of AI policy.

past the boredom of the plenary, we need dynamic formats: • Policy Sandboxing Live: Interactive sessions where policymakers and developers test-run a proposed regulation against a hypothetical AI product to see where it breaks. • The Reverse Town Hall: Instead of experts lecturing the public, youth leaders and marginalized community members present their AI pain points to a panel of developers and ministers. • Digital Twins for Policy: Using a simplified AI model to simulate the socio-economic impact of a proposed governance rule in real-time during the Dialogue. • AI Art & storytelling gazes: Using creative media to visualize AI futures, making abstract governance concepts feel real and urgent to non-technical participants.

From my experience, the most effective governance isn't just a set of rules; it's a living ecosystem of tools and policies that bridge the gap between innovation and safety. Here are four models that are currently moving the needle: • Risk-Based Statutory Frameworks such as Kenya's AI Bill 2026: Kenya is positioning itself as a leader by adopting a tiered risk model. By categorizing AI into "Unacceptable," "High," "Limited," and "Minimal" risk, the bill ensures that we aren't over-regulating harmless chatbots while keeping a strict eye on high-stakes areas like education and law enforcement. This nuanced approach is a great example of protecting the public without suffocating the local startup scene in Kilimani. • Regional Sovereign AI Strategies including The AU Continental AI Strategy: This is a powerful shift from being passive consumers to active creators. By focusing on regional compute hubs and green AI powered by renewables, the African Union is showing how governance can tackle the compute divide while ensuring our diverse linguistic and cultural data remains under our own sovereignty. • Regulatory Sandboxes such as EU AI Act Implementation: These safe spaces allow startups to test-drive their AI models under the guidance of regulators before going to market. It's a brilliant way to provide legal certainty for SMEs who might otherwise be priced out by compliance costs. • Automated Governance Platforms include NIST AI RMF & ISO 42001: Organizations are now moving from static policy documents to active operations. Using frameworks like the NIST AI Risk Management Framework, teams can embed ethics directly into the code-testing for bias and hallucinations in real-time rather than as an afterthought.