N/A

First, regulatory convergence on preventing the development of misaligned AI. This means establishing shared reporting requirements for large AI training runs, and agreed-upon triggers for mandatory pauses when models demonstrate signs of misalignment during development. Alongside this, there needs to be an expectation of greater due diligence from private actors particularly those providing compute infrastructure such as supercomputers and cloud services to understand and take responsibility for how their services are being used in AI development. Second, meaningful dialogue between AI-developing states and regions where significant AI training and data annotation work is being conducted. Countries such as Kenya host large data labelling workforces that are central to how frontier models are built, yet have little visibility or influence over the governance guardrails applied to those processes. The Dialogue should produce commitments to coordinate on standards for how AI training is conducted across jurisdictions, not just where companies are headquartered. Third, concrete investment conversations tied to bridging the AI divide. Regions with demonstrable potential to host data centres areas with available land, renewable energy capacity, and competitive connectivity should be actively included in infrastructure planning discussions. Success here would mean not just acknowledging this potential but beginning to translate it into investment pathways that allow these regions to participate in AI development rather than simply absorbing its consequences. This is especially important as AI systems increasingly displace some of the low-skill labour advantages that Global South economies have historically offered.

Social, economic, ethical, cultural, linguistic and technical implications of AI;Interoperability of governance approaches;AI capacity-building;Safe, secure and trustworthy AI;

Safe, secure and trustworthy AI It is critical that AI systems, in their use and in the outputs they generate, remain aligned with human values and preferences. Continuous monitoring and evaluation frameworks are needed to ensure that AI contributes to societal benefit rather than harm. Without this foundation, the other areas of governance lack meaning. AI capacity-building Increasing the level of informed public participation in AI governance discussions is essential for ensuring that regulatory frameworks genuinely reflect the interests of affected communities. Capacity-building enables more democratic engagement with AI systems and reduces the risk of governance being captured by narrow technical or commercial interests. Social, economic, ethical, cultural, linguistic and technical implications of AI This priority speaks directly to the experience of countries like Kenya, where AI deployment is already affecting cultural practices, language representation, and economic structures. Concerns include the erosion of low-resource languages in AI systems, the spread of disinformation, the economic effects of automation on workers, and the potential for AI to be leveraged for manipulation and political harm. Interoperability of governance approaches Given the significant difference in interests between AI-developing states and AI-using states, interoperability is essential to ensure that governance frameworks reflect all parties' priorities. A more interoperable governance landscape would help balance innovation incentives with protection obligations, and reduce the risk of regulatory arbitrage where companies exploit favourable conditions in less-regulated jurisdictions.

An area requiring greater attention is the push for context-specific regulation. States are at very different stages of AI deployment, and their regulatory needs differ accordingly. Encouraging countries to develop their own contextually grounded regulatory approaches rather than simply adopting frameworks designed for more advanced AI economies would help produce governance that is fit for purpose. This also has a collective benefit: where countries in similar contexts align their regulatory approaches, they create a regional floor of protection that reduces the ability of companies to move between jurisdictions in search of the most permissive conditions. Without this, the risk of a race to the bottom in AI governance is significant, particularly in Africa and other regions eager to attract AI investment.

Capacity-building gaps The absence of meaningful AI capacity-building has resulted in very limited public engagement in AI safety discussions in Kenya and the broader East African region. Without informed participation, regulatory conversations have been thin, and adoption of AI tools in sectors such as healthcare, education, and public administration has proceeded without adequate critique or community input. The result is deployment that may not reflect local needs or values. Social, cultural, linguistic and technical implications AI systems have largely been built without adequate training data for Kenyan languages and cultural contexts. Where AI is deployed in education, cultural and linguistic specificity is often absent, displacing local knowledge and practices. Beyond culture, the region has also experienced the spread of AI-enabled disinformation, including in political contexts, with limited mechanisms to respond. There are also growing concerns about the use of AI tools by authoritarian actors and the threats AI-enabled cyberattacks pose to national infrastructure. Interoperability gaps Current AI governance frameworks are largely designed by and for developed-state jurisdictions. They lack the nuance required to respond to the circumstances of countries like Kenya, including informality in the economy, limited institutional enforcement capacity, and specific vulnerabilities in labour markets and public services. The absence of interoperability means Kenya's interests are rarely reflected in the rules that ultimately govern the AI systems it uses.

On interoperability, the Dialogue can facilitate the development of mutual recognition frameworks and shared baseline standards that allow different regulatory approaches to coexist without creating governance vacuums. This is especially important for regions like East Africa where domestic regulatory capacity is still developing — shared frameworks reduce the burden on individual countries while ensuring basic protections apply. On AI training coordination, the Dialogue can promote bilateral and multilateral agreements between AI-developing states and jurisdictions where significant data labelling and training work is conducted. Countries like Kenya are already embedded in global AI supply chains through their data annotation workforces, yet have no formal seat at the table when training standards and oversight mechanisms are designed. The Dialogue can create the expectation that such coordination is normal and necessary, and can surface mechanisms for reporting, oversight and accountability that apply across borders. In doing so, the Dialogue would begin to address one of the more consequential governance blind spots in current frameworks: that global AI safety is only as strong as the least-governed jurisdiction in the training chain.

Regional bodies offer some of the most contextually grounded starting points. The African Union's AI continental strategy and the work of organisations such as the Smart Africa Alliance reflect an understanding of AI governance that is sensitive to the specific conditions, infrastructure constraints, and developmental priorities of African states. The Dialogue should actively partner with and build upon these bodies rather than positioning global mechanisms as replacements. At the level of bilateral and multilateral agreements between states and private AI companies, there is a need for the Dialogue to pay attention to the terms on which private actors negotiate access to compute resources, data, and markets in developing countries. These negotiations frequently occur without adequate public scrutiny or representation of affected communities. The Dialogue can play a convening role supporting regional bodies and national governments with tools and frameworks to participate more effectively in these negotiations, and establishing expectations around transparency and accountability. The added value of the Dialogue lies precisely in its ability to bring together these regional and national processes under a common framework that creates accountability across jurisdictions, without displacing the contextual knowledge that regional bodies carry.

Inter-agency panels that bring together technical and governance expertise in the same space, avoiding the tendency for technical and policy discussions to proceed in silos. Panels that represent states at different phases of AI deployment including both frontier AI developers and countries primarily using or hosting AI infrastructure so that the full range of interests is visible. Comparative regulatory panels that explore different models, such as process-based versus outcome-based regulation, and state-led versus private sector-led governance, drawing out what has worked and what has not in different contexts. Sessions featuring states that have risen significantly in AI readiness indices, sharing what enabled that progress and what it required in regulatory and investment terms. Roundtables that bring governments, private actors, and policy groups together to discuss what a regulatory and investment climate that serves both innovation and protection looks like in practice.

First, companies particularly small and medium-sized enterprises that have deeply integrated AI into their operations and are navigating its consequences without regulatory guidance or recourse. Their practical experience of what AI governance does and does not address is rarely heard in formal settings. Second, private citizens who are heavy users of AI systems and whose daily lives are shaped by algorithmic decisions. Their perspectives on trust, harm, utility, and fairness are the most direct evidence available about whether AI governance is working, yet they are almost never present in policy conversations. Third, individuals and communities who have experienced direct harm from AI systems whether through automated decisions in public services, exposure to AI-generated disinformation, or exploitation in data supply chains. These voices carry the clearest account of governance failure and should be structurally included in Dialogue proceedings, not as token representation but as a core input into agenda-setting.

Thematic working groups, open-floor civil society sessions, and live translation

Context-responsive, principle-focused, and agenda-oriented. This combination enables meaningful measurement of success and makes tradeoffs between competing objectives legible and manageable.