Cyril Shroff Centre for AI Law and Regulation (CSCAILR), O.P. Jindal Global University, Sonipat, India.

The first Global Dialogue on AI Governance would be a success if it produces outcomes that are concrete, inclusive, and capable of shaping future institutional action, rather than remaining at the level of general aspiration. First, success would mean the articulation of a shared set of minimum governance guardrails for AI development and deployment, particularly around safety, accountability, transparency, human oversight, and human rights. The Dialogue should clarify the baseline principles that no legitimate governance framework can ignore. Second, it would also require meaningful inclusion of perspectives that have often been peripheral in global AI governance discussions, especially those from the Global South. A credible Dialogue must ensure that those voices influence the relevant priorities, framing, outcomes etc. Third, the Dialogue should create a pathway for continued cooperation. This could include mechanisms for exchange of regulatory practices, a roadmap toward common evaluation and audit standards, among other things. Without continuity, even a well-conducted first meeting will have limited value. Finally, success would also mean recognising that AI governance is not only about managing downstream harms, but also about addressing deeper structural issues such as concentration of power, institutional capacity, and unequal access to technical infrastructure. In that sense, the true measure of success is whether the Dialogue helps move the international community from broad concern to a more coordinated, legitimate, and action-oriented governance process.

• Safe, secure and trustworthy AI
• Transparency, accountability, and human oversight
• Protection and promotion of human rights
• Open-source software, open data and open AI models

The four thematic areas we have selected reflect what we consider the irreducible minimum for a governance framework that is both coherent and legitimate. Safe, secure and trustworthy AI is foundational. No governance architecture is meaningful if baseline safety guarantees are absent. Trustworthiness must be treated as a substantive standard, and it must precede, not follow, large-scale deployment. In practice, it means building and deploying AI systems that behave predictably, can be audited, and can be meaningfully corrected when they cause harm. Transparency, accountability, and human oversight are the mechanisms through which safety commitments are enforced in practice. Without them, even well-designed regulatory frameworks remain unverifiable. Accountability and human oversight require visible decision trails, clear lines of responsibility, and genuine human capacity to intervene when systems fail. Human rights protection and promotion should anchor the entire governance enterprise in its proper normative foundation. AI systems are not ends in themselves. Their legitimacy depends on how they affect individuals and communities. For that reason, a rights-based approach must be embedded in governance design from the outset, rather than added later as an afterthought. Lastly, open-source software, open data, and open AI models respond to a structural inequality that could otherwise undermine the legitimacy of emerging governance frameworks. Countries and institutions without meaningful access to foundational models, datasets, and technical infrastructure are likely to remain merely adopters of technology rather than builders. In this context, openness is a condition for more inclusive participation in the development, evaluation, and governance of AI systems. These four areas are, in our view, deeply interconnected. CSCAILR's research on AI regulation, algorithmic accountability, rights implications of AI-driven decision-making among other fields, in the Indian and Global South context informs each of these priorities.

Yes. At least three cross-cutting issues deserve explicit recognition. First, institutional capacity and regulatory coordination. Even the best substantive principles will remain aspirational unless public institutions have the expertise, resources, and procedural mechanisms to implement them. This includes sector-specific regulators, standards bodies, procurement authorities, and courts. In many jurisdictions, especially in the Global South, the central challenge is not only what the rules should be, but who will enforce them, with what competence, and through what forms of coordination. Second, concentration of power across the AI value chain. This issue cuts across safety, accountability, rights, and openness. Control over compute, cloud infrastructure, foundation models, data pipelines, and technical talent is increasingly concentrated in a small number of firms and states. That concentration can distort markets, limit meaningful participation, and reduce the policy autonomy of less technologically advanced jurisdictions. Governance must therefore address not only downstream harms, but also upstream structural dependency. Third, labour and environmental impacts. AI governance is often framed in terms of users, risks, and systems, but less attention is paid to the workers and material infrastructures that make AI possible. Questions of data labour, content moderation, invisible annotation work, energy use, water consumption, and supply-chain extraction are becoming increasingly important. These concerns do not sit neatly within a single theme, but they affect the legitimacy of the overall governance project. A further emerging issue is epistemic integrity, meaning the effect of AI on shared knowledge, public reasoning, and informational trust. As generative systems increasingly shape what people see, believe, and rely upon, governance must attend not only to individual harms but also to the conditions of collective democratic life. In that sense, capacity, concentration, labour and environmental justice, and epistemic integrity may be worth recognising as additional cross-cutting priorities.

Governance gaps and related developments significantly impact India as a jurisdiction. While India is rapidly adopting AI in public services and the private sector, regulatory and institutional frameworks are still evolving. First, in the area of safe, secure, and trustworthy AI, a major governance gap is the absence of a single comprehensive law. Instead, India relies on adapting existing laws, which can lead to fragmented oversight and inconsistent standards across sectors. This affects sectors such as finance, healthcare, and digital governance, where AI systems process sensitive data. Weak and non-uniform security controls may increase risks of deepfakes, misinformation, and cyberattacks. Second, regarding transparency, accountability, and human oversight, governance gaps are evident in unclear liability and decision-making responsibility when AI systems fail or produce biased outcomes. Lack of transparency in algorithmic processes can undermine public trust and make it difficult to challenge AI-based decisions. In India, different government departments often deploy AI tools independently without uniform accountability mechanisms, thereby creating governance inconsistencies. Third, in the area of protection and promotion of human rights, rapid AI deployment has raised concerns regarding surveillance and discrimination. Large-scale data collection and automated decision-making can threaten individual dignity, especially when safeguards are weak. Also, India's recognition of privacy as a fundamental right has increased pressure on policymakers to ensure that AI systems respect constitutional protections and human rights standards. Finally, developments in open-source software, open data, and open AI models offer opportunities for innovation and transparency but also create governance challenges related to data quality, intellectual property, and misuse of open technologies. Open AI ecosystems require strong standards and coordination across agencies to ensure responsible development and public trust. Overall, governance gaps challenge responsible adoption of AI and increase regulatory uncertainty. This highlights the need for uniform legal frameworks, institutional coordination, and human-centred oversight.

Many countries are failing to effectively regulate AI out of concerns that regulation will lower the pace of progress relative to countries which do not regulate the field. Thus, even if risks and possibilities of harm are evident, countries are shying away from active governance of AI. International cooperation toward common guardrails and global standards that create a level playing field is the only solution to this issue. The AI Dialogue is appropriately positioned to advocate for, and forge, a multilateral framework of common, minimum guardrails and standards for AI development and deployment. The AI Dialogue should attempt to create consensus among major jurisdictions to initiate concrete and time-bound negotiations for a multilateral framework for AI governance. The AI Dialogue may also attempt to crystallize general principles which would form the basis for such a multilateral AI governance framework. In doing so, the AI Dialogue can align and map AI governance with broader UN goals, including protection of human rights and Sustainable Development Goals (SDGs). A less ambitious but useful contribution of the AI Dialogue could be in creating a pathway for regular and quick exchange of information and regulatory best practices between UN member states, with each member nominating a nodal agency or official for this purpose. This could be more formal than the existing OECD.AI Policy Observatory. The AI Dialogue should also try and work toward common AI evaluation and audit standards.

The OECD has been working in the field of AI governance for a while now, and the OECD AI Principles is a good starting point while looking at inter-governmental or multilateral standards for AI. The AI Dialogue can build upon these principles and explore how they can be made workable, equitable, and implementable. The AI Dialogue is well positioned to bring together diverse opinions from various jurisdictions and ensure that AI governance principles are equitable and representative of a wider range of interests. Particularly, the AI Dialogue can bring forward the views and concerns of the Global South and ensure that AI governance principles do not leave out their perspective. OECD's Global Partnership on AI (GPAI) has had some success in bringing together countries for multilateral and multi-stakeholder deliberations on AI governance. The AI Dialogue can benefit from the experience of, and the framework created by, the GPAI and should consider collaborating with the initiative to jointly explore AI governance best practices and challenges. The AI Dialogue should also utilize the expertise of the Independent International Scientific Panel on AI established in 2025 under the aegis of the U.N. and let the panel provide scientific insights and datapoints which could set the tone for the discussions to follow.

Events focused around creating a dialogue on AI typically feature panel discussions, release of reports, and presentations of work done by various stakeholders. While they are valuable, the real challenge is what happens after the conversation ends. Too often, rich multi-stakeholder discussions do not translate into concrete policy outcomes or governance frameworks. The Dialogue must be designed with follow-through built in from the start, not treated as an afterthought. Discussions should feed into clearly identified policy processes, with timelines, responsible actors, and review mechanisms that are agreed upon before the Dialogue concludes. On structure, we would recommend a tiered approach. Smaller working group sessions, organized by theme, should do the substantive analytical work and produce draft recommendations. These then should feed into a larger plenary where broader political endorsement and visibility can be achieved. This prevents the plenary from becoming a stage for prepared statements rather than a genuine exchange. Different stakeholders contribute best when their roles are clearly defined. Governments bring regulatory authority and implementation experience. Academic and civil society institutions bring independent analysis and the ability to ask uncomfortable questions. The private sector holds technical knowledge that is necessary for informed conversations, but its participation should be structured so that it informs the process rather than shapes its conclusions. Pre-Dialogue consultations, particularly with researchers and institutions from the Global South, should be part of the formal preparatory process. The Global South remains underrepresented both in the datasets that underpin AI systems and in governance conversations that shape how those systems are regulated. Cultural and social contexts vary significantly across regions, and governance frameworks that do not account for this risk being both technically inadequate and democratically illegitimate. Written input mechanisms, like this one, must be demonstrably connected to the actual agenda and post-event outcomes to remain meaningful.

The most significant and apparent gap in global AI governance discussions is not merely geographical but also structural. The communities most consequentially affected by AI systems, such as the ones subject to automated decision-making or impacted by its infrastructural or environmental consequences, are almost entirely absent from the rooms where governance frameworks are discussed and designed. Within this broader gap, several specific absences stand out. Researchers and institutions from the Global South are routinely engaged as recipients of governance frameworks developed elsewhere rather than as co-architects of those frameworks. This is particularly acute for South Asia, where AI deployment is accelerating rapidly but domestic regulatory scholarship remains under-resourced and underheard at the international level. Legal and regulatory scholars, as distinct from computer scientists and economists, remain underrepresented even though the core questions of AI governance are fundamentally questions of law, accountability, and institutional design. Indigenous communities, whose data, cultural knowledge, and ways of life are increasingly implicated in AI systems, have almost no formal presence in these conversations. Inclusion requires resources for participation, genuine agenda influence, and follow-through that connects these voices to actual governance outcomes. Pre-Dialogue regional consultations, dedicated track funding for Global South institutions, and formal mechanisms for academic and civil society input to feed into negotiating the framework's text would be meaningful first steps.

A meaningful and dynamic AI Dialogue will require formats that move beyond long prepared statements and create genuine exchange across governments, industry, academia, civil society, and communities most affected by AI systems. The most effective design would combine structured dialogue with participatory and problem-solving formats. The dialogue should include thematic roundtables with mixed stakeholder seating, rather than sector-based clustering. This would allow a regulator, engineer, academic, private-sector representative, and civil society actor to respond to the same question in the same room. The discussion can also benefit from scenario-based policy labs. Instead of discussing AI governance only in the abstract, participants could work through realistic cases such as cross-border deepfake harms, AI use in public services, compute concentration, or child safety risks.

Several existing models already offer concrete lessons for effective AI governance. The European Union's AI Act is especially important because it moves beyond broad ethical commitments to a risk-based legal framework. It prohibits certain unacceptable uses, imposes heightened obligations on high-risk systems, and avoids treating all AI systems alike. That is a useful example for global governance. It shows that innovation and accountability do not have to be framed as competing goals, and that regulation can be calibrated to actual levels of risk. At the international level, UNESCO's Recommendation on the Ethics of Artificial Intelligence and its Readiness Assessment Methodology (RAM) provides a strong model for countries that are still developing institutional and regulatory capacity. The Recommendation anchors governance in human rights, dignity, transparency, fairness, and human oversight, while the RAM translates those principles into a practical diagnostic tool across legal, social, educational, economic, and technical dimensions. This is particularly valuable for ensuring that global AI governance remains inclusive and does not privilege only the most technologically advanced states.