DAVID NEE NGOYA TSIMI

A successful first Global Dialogue on AI is very important for the future. Governance should create clear and understandable next steps; not just formal talk. Firstly , it would be good if countries can agree on basic principles, like AI should be safe, transparent, fair, responsible and people should stay in control. We have already seen problems for example, during recent elections in different countries, AI tools were used to create fake images and videos (deepfakes) that spread false information online. This shows why basic rules are needed. Secondly , the dialogue should help build trust. It should lead to cooperation between countries, researchers and companies to be able to share information about AI risks, evaluation methods and best practices . Some large AI companies test their systems before release, but they do not always share full results. In the past, lack of transparency in tech (like with social media data misuse cases) caused harm, so more openness would help. Thirdly , there should be real actions, not just discussion. Countries could agree that strong AI systems must be tested before use. A real example is how the European Union introduced the AI Act to regulate high-risk AI systems before they are widely used. This shows rules can be created early. It is also important to include different countries, not only powerful ones, because AI affects everyone. Finally, there should be a clear next step, like future meetings or actions to take. Success means real agreement, trust, and clear actions to keep AI safe.

• Safe, secure and trustworthy AI
• Protection and promotion of human rights
• Transparency, accountability, and human oversight
• Social, economic, ethical, cultural, linguistic and technical implications of AI

Firstly, I choose secure and trustworthy AI because AI is quickly becoming very powerful. If it is not properly tested and controlled, it can cause serious harm, such as spreading false information, making unsafe decisions, or being misused by bad actors. There is a strong need to act now before these risks grow. For social, economic, ethical, cultural, linguistic and technical implications of AI, we can see that AI is already changing jobs, education, and daily life. If we do not manage this change, it could increase inequality, replace jobs without support, and ignore some cultures or languages. Also protection and promotion of human rights is very important as AI can threaten basic rights like privacy, freedom of expression, and equality. For example, AI can be used for surveillance or discrimination. Strong action is needed to prevent abuse. Lastly transparency, accountability, and human oversight because many AI systems work like "black boxes." If we do not understand how they make decisions, it is hard to trust them or hold anyone responsible. Humans must stay in control, especially in important decisions.

Yes, I think there are a few important issues that are not fully covered in the list. First, AI is developing very fast, and this really changes how I see governance. Rules often take time to agree on and apply, but AI is improving much quicker than that. This makes me worry that policies could always be one step behind the technology. So I think systems need to be more flexible and updated regularly. Second, I am concerned about unequal power between countries and companies. Right now, only a few big actors have access to the most advanced AI. This could create a situation where a small group has too much control over technology that affects everyone. I think this makes global cooperation even more important. Third, I think measuring what AI can really do is still a big problem. We often hear that AI is "advanced," but it is not always clear what that means in real life. Without proper testing and shared standards, it is hard to judge risks properly or make fair rules. In simple terms, from my perspective, AI governance needs to deal with how fast AI is changing, who controls it, and how we properly measure its real abilities.

In the legal sector, and from my focus area of AI governance and regulation, I see clear gaps that are already shaping how AI is used in legal work. One major challenge is responsibility and liability. For example, if a lawyer uses an AI tool to draft a contract or prepare a legal argument and the AI includes a wrong case or false citation, it is still unclear who is legally responsible. From a governance point of view, this shows a gap in regulation on accountability for AI-assisted legal work. Another key issue is confidentiality and data protection. In practice, lawyers may use AI tools to summarise case files or prepare documents. But if these systems are not properly regulated, sensitive client data could be stored or processed outside secure legal systems. This raises the need for stronger regulatory rules on how AI tools can handle legal data. There is also the issue of bias and fairness. For example, if AI systems are used to support case analysis or predict outcomes, they may reflect biased patterns from past legal decisions. From a regulatory perspective, this raises the need for standards on testing and auditing AI systems used in legal contexts. At the same time, there are important opportunities for AI governance. In my research area, I see how regulators could introduce mandatory transparency rules for legal AI tools, such as requiring disclosure when AI is used in drafting or advising. Another example is pre-deployment testing requirements, where AI systems used in legal settings must be tested for accuracy and bias before approval.

From my perspective, the AI Dialogue can play a very important role in turning scattered national efforts into real international cooperation on AI governance. I am actually writing a paper in this direction, focusing on how global coordination can make AI regulation more effective. I think firstly that the Dialogue can help create a shared understanding of key risks and priorities. Right now, countries describe AI safety in different ways. A global discussion can help align these views so that terms like "safe AI," "high-risk systems," and "responsible use" are understood more consistently. This is important for my research, because regulation only works well when there is a common language. The dialogue can build trust between governments. One of the biggest problems in AI governance is not only technical, but political—countries worry about falling behind or losing control. A regular forum for sharing concerns, incidents, and best practices can reduce this tension and encourage cooperation instead of competition. Also, it can support practical coordination. For example, the Dialogue could help countries agree on common standards for testing advanced AI systems before release, or for reporting system capabilities. This is especially important because AI tools are cross-border and cannot be regulated by one country alone. Finally, from my academic work on AI governance, I see the Dialogue as a way to connect research and policy. It helps ensure that regulatory ideas are based on real-world risks and not developed in isolation. I see the AI Dialogue as a bridge between countries, between research and policy, and between fast AI development and slower legal systems trying to regulate it.

From my perspective, the AI Dialogue should build on existing efforts rather than starting from zero, because there are already important initiatives in AI governance. For example, the OECD AI Principles and the G7 Hiroshima AI Process already provide shared ideas on trustworthy AI and risk management. These are useful because they show that some global agreement is already possible. However, they are still quite high-level, and often lack strong enforcement or practical testing standards. Another important mechanism is the EU AI Act, which is one of the most developed legal frameworks for AI regulation. It gives a risk-based approach, especially for high-risk systems. But it is regional, not global, which creates fragmentation. In my research area of AI governance, I often see this problem: rules are strong in one region but missing or different elsewhere. There are also technical and safety-focused initiatives, such as frontier AI safety evaluations by research organizations and companies, which test model capabilities and risks. These are very useful, but they are not always shared in a consistent or transparent way across the industry. I really think that the added value of the AI Dialogue would be to connect these separate efforts into a more coordinated global system. It could help align principles from OECD and G7, bring regulatory lessons from the EU AI Act into global discussion, and encourage shared standards for AI testing and reporting.

I will say that, different stakeholders should each have a clear and active role in the AI Dialogue, because AI governance cannot be shaped by governments alone. Governments should take the lead in setting direction and agreeing on shared priorities, especially around safety, regulation, and international coordination. However, they should also be open to input from others, because AI is developing faster than most legal systems can respond. Researchers and academic experts, including those working in AI governance like myself, should contribute evidence-based analysis. For example, studies on capability trends, risk assessment, and evaluation methods (such as time-horizon type research) can help ground discussions in real data rather than assumptions. The private sector should also play a central role, since AI development is mainly driven by companies. They can share technical insights, safety testing methods, and real-world deployment challenges. However, there should also be clear expectations for transparency, especially around high-risk systems. Civil society should represent public concerns, especially around rights, fairness, and accountability. This is important to ensure the dialogue is not only technical but also socially responsible. In terms of structure, I think the Dialogue should be regular, not one-time, with clear follow-up mechanisms. It should include both high-level policy discussions and technical working groups focused on specific issues like safety evaluation, transparency, and governance standards. There should also be space for regional representation so that smaller or developing countries are not left out

Just as I mentioned earlier, several important voices are still underrepresented, one key gap is developing countries and smaller states. Many global AI rules and standards are shaped by a few major powers, even though AI systems are used worldwide. This can create rules that do not fully reflect local needs or realities. These countries should be included through stronger regional representation, funding for participation, and dedicated seats in global forums. Another underrepresented group is civil society and ordinary users, especially people who are directly affected by AI in daily life. For example, workers whose jobs are changed by automation, or students using AI in education, are often not part of high-level policy discussions. They could be included through public consultations, citizen panels, and structured feedback processes. There is also a gap in technical voices from independent researchers and academia, especially those outside big companies. Many important insights on AI risks and governance come from research institutions, but they may have less influence compared to industry actors. This could be improved by creating stronger channels for academic input into policy decisions. In addition, legal professionals and regulators in smaller jurisdictions are often less visible in global debates, even though they deal with real-world AI legal issues. Their experience could be better integrated through cross-border legal working groups and training programs. Global AI governance can be improved by making participation more balanced, not only including powerful states and companies, but also smaller countries, everyday users, independent researchers, and legal practitioners who deal with AI's real impacts.

A useful format would be small, focused working groups alongside the main dialogue sessions. From my experience working in AI governance, I find that real progress often happens in smaller groups rather than only in large formal meetings. If governments, researchers, industry, and civil society work together in these groups on specific issues like AI safety testing, transparency, or liability, it would make the discussion more practical and easier to turn into real policy ideas. Another effective format would be live case studies and scenario exercises. I think this is especially important because AI governance can feel very abstract until you see real situations. For example, participants could work through a case where an AI system gives wrong legal advice or influences public information, and then discuss what rules or responses should apply. This helps expose gaps in current governance in a very direct way. Finally, I would strongly support continuous digital participation platforms. In my view, AI governance cannot only happen during official meetings because AI is changing too quickly. An ongoing online space where experts, policymakers, and even the public can share evidence, raise concerns, and comment on proposals would keep the Dialogue active and relevant between sessions.

The EU AI Act is, in my view, one of the strongest real examples of AI governance. I find its risk-based approach very useful because it does not treat all AI systems the same. Instead, it focuses more strict rules on high-risk areas like hiring, education, and legal decision-making. From my perspective, this is important because it directly connects regulation to real harm in society. The OECD AI Principles are another framework I often refer to in my thinking about AI governance. They focus on transparency, fairness, accountability, and human control. Even though they are not legally binding, I see them as important because they create a shared language that many countries can agree on. In my view, this kind of soft law is often the first step toward stronger global cooperation. I also find AI safety testing and evaluation practices very important in real implementation. These are used by researchers and companies to test models before release and check for risks like bias, unsafe outputs, or misuse. At the same time, I see value in multi-stakeholder platforms like the G7 Hiroshima AI Process, where governments and industry meet to discuss common standards. I will say, combining regulation, technical testing, and international dialogue is the most realistic way to make AI governance effective