CODATA, the Committee on Data of the International Science Council

The first Global Dialogue on AI Governance should produce a shared agenda for action, along with the development of general principles. A key action item of the Dialogue should identify mechanisms for ongoing collaboration and the specificities of those mechanisms. Success would mean establishing shared missions, visions, trust, and action items. This should build towards a framework that defines dimensions or categories needed for AI governance. There are so many disparate approaches, and many seem incomplete. An example might be UNESCO's Recommendation on Open Science that defined values, principles, and areas for action. Therefore, what would make the first Global Dialogue on AI Governance a success includes: (1) substantive consensus, defining and clarifying shared/common understandings on priority areas for international cooperation, while balancing against the recognition that there will be different national governance approaches; (2) a commitment to practical collaborations, through various strategic international partnerships that help close AI access and capacity/skills gaps and address the contextual realities that we live in a world where there is still major digital divides; and (3) effective coordination, developing a mechanism/s that connect existing efforts (there is no need to duplicate or reinvent the wheel) working from what is in place but acknowledging that the UN is in a unique position to strengthen commitments to human rights, transparency, accountability and robust human oversight. Measurable outcomes for success are: 1. A traceable roadmap 2. Clear definition of participants and their roles 3. Sustainable funding 4. Annual reporting 5. A pluralistic, participatory, and collaboration-based structure 6. Bias detection 7. Establishing a platform that ensures continuous, effective communication.

• Safe, secure and trustworthy AI
• Transparency, accountability, and human oversight
• AI capacity-building
• Open-source software, open data and open AI models

It is essential to avoid a deterministic perspective that suggests AI development is autonomous, leaving humans with no influence over its trajectory and forcing us to adapt to a paradigm fully designed by AI. On the contrary, we must act as active agents, critically reflecting on the social, economic, ethical, cultural, and technical implications of AI, and proactively shaping its course. The protection and promotion of human rights should be central to this trajectory, alongside efforts to prevent bias and the deepening of existing inequalities. Humans must not be treated as mere means to an end, reduced to data sources, or positioned as passive consumers of technology. In relation to data, the really important thing is scientific quality. Data, software and models should be open, yes, but open data that is high quality, traceable, transparent verifiable data is a priority! Additionally, the following three priorities are foundational to responsible and equitable AI governance. Safe, secure, and trustworthy AI is essential because AI systems increasingly shape all sectors of society. AI capacity-building is equally urgent because many institutions, countries, and communities lack the infrastructure, expertise, and policy support needed to participate meaningfully in AI development and governance. Transparency, accountability, and human oversight are necessary to ensure that AI systems remain understandable, safe, and accountable, serving humans and enhancing human activities. Although all the options are important, less priority should be put on "Interoperability of governance approaches".

Governance of the data ecosystem, which makes AI possible: data integrity, cultural sensitivity, data sovereignty is fundamental to AI governance. Also important are: Computing and infrastructure governance (linked to access, capacity/skills and digital divide), many are struggling to keep up with high-performance AI innovations, and at the same time how best to govern within limited resources context. Data rights and data integrity (linked to the theme on open data and open intelligence models as well as mis/disinformation). Development, deployment and use ( linked to model lifecycle oversight, especially concerning model and data drift in adaptive technologies). Environmental impact (energy, water, sustainability, risks and harms), these technologies/systems demand enormous resources.

Many education institutions are adopting AI tools faster than they can develop clear policies on privacy, data use, transparency, academic integrity, accessibility, and human oversight. This creates uncertainty for educators, students, researchers, and administrators. The challenge is to move beyond reactive policies toward coherent, evidence-informed governance that protects learners and researchers while enabling responsible innovation, experimentation, and capacity-building. The health sector is one of the pioneering fields in the use and development of AI. However, the lack of legal and ethical regulations creates complexities. Data access remains a significant problem for AI development. These challenges are also valid for scientific research and academic publishing. This era is reminiscent of the early days of the web, when people were adopting it as fast as possible without fully understanding the implications. Because AI is so empowering, individuals can initiate great action on their own without help. We have yet to understand all of the ramifications, security issues, data disclosures, and resultant misuses currently underway. There will be a huge catch-up and correction needed once we start to see serious misuse of data shared with AI and/or security exploits in AI technology. The use of AI for misinformation, including by governments is a major issue. Compliance by large AI corporations and adherence to guidelines for safe, trustworthy use is essential. The large businesses that invest the most in AI do not follow the same best practices and rules that are adopted by academia and organizations in more regulated countries. In some countries, there remains a major governance gap concerns limited regulatory capacity, technical evaluation, and enforceability mechanisms are seriously lagging, AI governance risks becoming more about 'paper compliance' rather than realised protections, undermining trust, widening divides, and weakening any attempts to keep up with the rapid advancements in AI.

The AI Dialogue has great potential as a continued platform to advance international cooperation, share best practices, and enable open, inclusive discussions to support SDGs and close divides as per the A/RES/79/325. The UN AI Dialogue can lend legitimacy and align standards to guide countries' interoperability approaches. The Dialogue can support international cooperation by sharing evidence, policy models, governance tools, and lessons learned from implementation. Its added value should be practical, serving as a pivot for sharing and envisioning AI systems and human-AI collaboration. The UNESCO-CODATA Data Policies for Times of Crisis Facilitated by Open Science (DPTC) toolkit and its checklist state that 'AI should not be actively implemented without rigorous evaluation and clear safeguards'. To implement this, the UN Global Dialogue on AI Governance process currently being developed provides a valuable opportunity to support international cooperation on AI governance to address current and real-time needs. This means sharing best practices, continuous and effective counseling on emerging issues, and collaboration in the development of guidelines and data policies. For those in the US, the global dialogue will provide a forum where topics prohibited by the US government (and where industry is not interested) can be advanced.

Relevant mechanisms include UNESCO's AI ethics recommendations, OECD AI principles, the Global Digital Compact, open science initiatives, FAIR data communities, and multi-stakeholder networks focused on data policy and research infrastructure. CODATA's work on data policy, data stewardship, and international scientific collaboration is also relevant. The added value of the Dialogue would be to connect these efforts, reduce duplication, and translate high-level principles into practical governance practices across regions and sectors. Although there are some international, bilateral, or national initiatives, they still lack the capacity to bring together effective partners and foster collaboration. CODATA has high potential to lead/ coordinate or participate in AI dialogue; there is still room for it to assume a more effective role on a global scale. Applying consistent laws and guidelines across countries, based on the UN guidelines (in a similar way that the UNESCO open science guidelines have helped with adoption). Ensuring equity of AI production (avoiding all the development in only a couple of countries) and access. Existing mechanisms, partnerships and initiatives to be built on include: OECD AI Principles, which are recognised as the first intergovernmental standard for AI; ISO/IEC 42001:2023 as the first international standard providing requirements for an Artificial Intelligence Management System (AIMS); UNESCO Recommendation on the Ethics of AI (2021) as the first global ethics standard focussed on human rights, transparency, fairness, oversight for common good.

The Dialogue could produce concise action briefs, implementation toolkits, and mechanisms for continued engagement so that participation leads to concrete outcomes rather than one-time discussions. Important stakeholders include governments, academia, the private sector, civil society, international organisations, technical experts, ethics committees, and lawyers. The format could be semi-structured multistakeholder roundtables; establishing ad hoc working groups (for emerging issues, persistent unsolved problems, or sporadic cases); open consultation phases; ethics review panels; capacity-building seminars; and streamlining these activities on a digital platform that enables effective participation. However, it will be essential to enable new experts to emerge and leapfrog others based on new, relevant expertise. There needs to be room for new voices. The "old guard" may not have the best idea of what is happening. Comparison with the WTO approaches to developing the TRIPS agreement (Trade-Related Aspects of Intellectual Property Rights) is instructive. TRIPS ended up being controversial because efforts were driven largely by developed nations pushing for high protection mechanisms of intellectual property, which developing countries could not later access (a gross misalignment with the needs of developing nations). UN AI Dialogue presents a vital opportunity for stakeholder contributions to include, for example, government, academia, civil rights society, private sector. This could take place through the seven themed tracks in moderated panel/discussion sessions, capacity-building efforts that include dialogue with diverse groups, and regional-level dialogue/engagement.

It is essential to involve the 'Global South' in general. More specifically, disadvantaged populations, elderly people, and people with low literacy, regardless of their country or region. Local communities that represent the indigenous in terms of linguistic and cultural context, representatives from the informal sector, youth and those who have disabilities.

Formats that allow stakeholders to share and examine concrete opportunities and challenges would be ideal. The AI Dialogue could use scenario-based policy labs, regional listening circles, participatory foresight workshops, and governance simulation exercises. Such formats would promote "specifics" and real-world experience, rather than discussing principles only in the abstract. For example, participants could work through cases involving AI in education, health, public services, research, or labor to identify governance gaps and practical safeguards. It is essential to engage with social and humanities scientists, negotiation experts, ethnographers, and other facilitation professionals when designing formats. At the same time, modalities will need to be grounded somewhat in accepted formats so as not to scare the audience off. For example, it can still be difficult to attract senior scientists to an unconference format event. Speaker rotation (don't only focus on the most vocal or confident voices (they do not always speak for everyone), draw out key voices); session findings could be used for policy or media briefs. Remember, policies and politics are two different animals, and sometimes they require some masters in diplomacy to navigate how to leverage one to get around the other.

In education and research, useful practices include syllabus-level AI disclosure expectations, privacy-first tool vetting, human review of AI-generated outputs, skepticism toward unreliable AI detection tools, and training for responsible AI use. Broader approaches include impact assessments, model documentation, data provenance standards, audit trails, and FAIR data practices. Platforms and policies should make AI systems explainable, contestable, and accountable while supporting innovation, openness, accessibility, and public trust. So much of AI use is hidden, since much of the power lies with the individual. Therefore, participatory practices, such as creating contributor covenants (as seen in open source software) may be effective in getting established groups to embrace new practices. This should be paired with active engagement plans to capture evolving practices, use cases, and to bring those lagging along so they are not left behind. Examples include: OECD AI Principles, UNESCO AI Ethics Recommendation, NIST AI Risk Management Framework (AI RMF), institutional ethics committees, and institutional ICT/ITS Steering committees. I do not think good governance is a 'one-size-fits-all' approach. We need to develop a comprehensive, context-specific, integrated response to AI governance, including regulatory frameworks, policies, practices, oversight, auditing, and tracking mechanisms.