GD GOENKA UNIVERSITY, INDIA

The first Global Dialogue will succeed if it moves past broad principles and produces at least three concrete results. First, the Dialogue should establish a working definition of what "interoperability" means in practice for AI governance. Right now, the EU AI Act classifies risk by system type. India's AI Governance Guidelines (released February 2026) classify risk by social context. The US has no federal AI law at all. These are not just different speeds of regulation. They are different logics. Unless the Dialogue produces a shared vocabulary for comparing these frameworks, interoperability will remain a slogan. A practical starting point would be a comparative mapping exercise across five or six national approaches, identifying where they overlap and where they genuinely conflict. Second, the Dialogue should foreground intellectual property as a governance issue, not just an innovation issue. Generative AI training on copyrighted works is already creating legal disputes worldwide, from the New York Times litigation in the US to India's DPIIT Committee examining copyright and AI training. Yet most global AI governance conversations treat IP as a side topic. If the Dialogue ignores this, it ignores one of the few areas where developing countries have direct legal leverage over how AI systems are built. Third, the Dialogue should create a formal channel for university-based policy labs from developing countries to contribute ongoing research to the Scientific Panel. Today, the institutions shaping global AI governance evidence are overwhelmingly located in North America and Western Europe. India alone has over 1,600 universities, many now building AI governance capacity. A structured fellowship or submission track would bring this work into the system rather than leaving it at the margins. Success is not a declaration. It is a mechanism that outlasts the event.

• Interoperability of governance approaches
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Transparency, accountability, and human oversight
• Open-source software, open data and open AI models

I lead a university-based AI Policy Lab in India. My selection reflects what we encounter on the ground. Interoperability is the most pressing structural gap. India released its AI Governance Guidelines in February 2026, choosing a context-sensitive, voluntary compliance model. The EU chose binding horizontal regulation through the AI Act. These approaches are not inherently incompatible, but without deliberate bridging work, they will fragment the global AI market in ways that hurt smaller economies most. A researcher in India who builds an AI tool for agriculture should not need separate compliance architectures for every jurisdiction. The Dialogue should prioritise practical interoperability, not just mutual recognition in principle. The social, economic, and cultural implications theme matters because AI governance discussions still underweight non-English linguistic contexts. India has 22 scheduled languages and hundreds of dialects. When large language models are trained primarily on English-language data, they carry embedded cultural assumptions into healthcare, education, and legal services in countries where those assumptions do not hold. This is not a hypothetical risk. It is a current deployment reality. Transparency and accountability are foundational. India's DPIIT committee is currently examining whether AI training on copyrighted works should require disclosure. My own research on copyright and generative AI shows that without mandatory transparency about training data, creators in developing countries have no way to enforce their rights. Accountability without transparency is empty. Finally, open-source AI models matter because they are the primary pathway through which developing countries will build sovereign AI capacity. India's IndiaAI Mission has already made over 38,000 GPUs available at subsidised rates and supports domestic foundation models. But open models need open governance norms too, or we risk replicating proprietary dependencies under an open label.

Three issues are missing from the current thematic framework. First, copyright and AI training data governance. This is not a niche intellectual property question. It is a structural power question. Companies in a handful of countries are training foundation models on the creative and cultural output of the entire world, often without consent, compensation, or even disclosure. India's Copyright Act of 1957 limits fair dealing to non-commercial research and does not clearly cover large-scale AI training. The EU carved out a text and data mining exception with an opt-out mechanism. Japan adopted a broad exception. These divergent approaches directly affect whose culture gets absorbed into AI systems and who benefits economically. The Dialogue cannot govern AI meaningfully without addressing the data supply chain. Second, AI governance capacity in universities. The current framework emphasises state capacity and infrastructure (compute, connectivity, skills). But the institutions that produce the next generation of AI policymakers, lawyers, and ethicists are universities. Most universities in the Global South lack dedicated AI governance research programs. Building a few centres of excellence is not enough. The Dialogue should encourage lightweight, replicable models. At GD Goenka University, we are building a zero-cost AI Policy Lab that runs on faculty expertise and student research rather than large grants. This model can scale across hundreds of universities in developing countries if there is institutional recognition and a knowledge-sharing platform. Third, the environmental costs of AI infrastructure. The race to build data centres and expand compute is accelerating across the Global South. India is actively inviting global AI companies to establish data centres domestically. But AI governance discussions rarely connect compute expansion to energy consumption, water usage, and carbon emissions. Any governance framework that promotes AI capacity building without addressing environmental sustainability is incomplete.

India is simultaneously building AI capacity at speed and governing it through a patchwork of existing laws that were not designed for this technology. This creates both real challenges and unusual opportunities. The biggest governance gap is the absence of a standalone AI law. India released its AI Governance Guidelines in February 2026, choosing a voluntary, "law-plus" model that layers guidance on top of the IT Act of 2000, the Digital Personal Data Protection Act of 2023, and sector-specific rules. This approach keeps compliance costs low for startups, but it leaves critical questions unresolved. For instance, the DPIIT committee examining copyright and AI training has not yet clarified whether large-scale training on copyrighted Indian works falls within fair dealing under Section 52 of the Copyright Act. Indian creators, publishers, and musicians are exposed to extraction of their work without a clear legal remedy. This is not an abstract concern. It affects the economic interests of one of the world's largest creative economies. On interoperability, the challenge is practical. India's context-sensitive risk classification does not map neatly onto the EU AI Act's system-based risk tiers. Indian AI companies building products for export face compliance uncertainty. Without a bridging framework, smaller firms in developing countries bear disproportionate costs. In the legal education sector, the gap is stark. India has over 1,600 universities, but almost none have dedicated AI governance research programs. Law students graduate without training in algorithmic accountability, AI ethics, or data governance. This means the next generation of Indian lawyers, judges, and policymakers will encounter AI disputes without adequate preparation. At GD Goenka University, we are addressing this by building an AI Policy Lab and embedding AI governance modules into law courses. But this needs to become the norm, not the exception.

The AI Dialogue can play three roles that no other existing forum currently fills. First, it can serve as a translation layer between divergent national governance models. Today, the EU AI Act, India's voluntary AI Governance Guidelines, China's sector-specific regulations, and Brazil's emerging AI Bill all operate on different logics. The OECD and G7 processes have attempted alignment, but they exclude most of the Global South from the table. The AI Dialogue, by design, includes all 193 UN member states. It should use this universality not to produce another set of principles, but to build practical tools: a shared taxonomy of AI risk categories, a comparative matrix of national governance approaches, and model interoperability clauses that countries can adapt into bilateral and regional agreements. This is the unglamorous infrastructure work that actually enables cooperation. Second, the Dialogue can become the place where the Scientific Panel's findings are stress-tested against policy reality. The Panel will produce evidence-based reports. But evidence does not automatically become policy. The Dialogue should create structured sessions where policymakers from different legal traditions respond to the Panel's findings with concrete national examples. A finding on algorithmic bias means something different in India's welfare delivery system than in a European credit scoring context. Unless the Dialogue creates space for that kind of contextual exchange, the Panel's work will remain academic. Third, the Dialogue can formalize the role of universities and research institutions from developing countries in global AI governance. Right now, the evidence base for AI policy is overwhelmingly produced by institutions in North America and Europe. The Dialogue should establish a standing submission track for university-based policy research, particularly from countries that are large-scale AI adopters but not yet major contributors to the governance literature. India, Brazil, Indonesia, and Nigeria all fit this description.

The AI governance landscape already has several serious initiatives. The Dialogue should not duplicate them. It should connect them in ways they cannot connect themselves.The OECD AI Policy Observatory has built the most comprehensive comparative database of national AI policies. The Global Partnership on AI (GPAI), now merged with the OECD's AI work, has produced valuable expert reports on responsible AI. But both are limited by membership. Most developing countries are not OECD members and had limited voice in GPAI's working groups. The Dialogue should draw on OECD data and GPAI research as inputs, while ensuring that countries outside these clubs shape how the findings are interpreted and applied.The AI Safety Summit process, running from Bletchley Park through Seoul and Paris to the India AI Impact Summit in February 2026, has built momentum on frontier AI risk. The Delhi Declaration signed by 88 countries committed to inclusive AI development. But summit declarations do not create follow-through mechanisms. The Dialogue can serve as the standing venue where summit commitments are tracked, reviewed, and connected to national implementation.UNESCO's Recommendation on the Ethics of AI, adopted by 194 member states in 2021, remains the only near-universal normative framework on AI. Yet implementation has been uneven and under-monitored. The Dialogue should request UNESCO to present periodic implementation assessments, turning a static recommendation into a living governance tool.At the regional level, the African Union's Continental AI Strategy and India's AI Governance Guidelines represent governance approaches built outside the transatlantic corridor. The Dialogue's unique added value is that it is the only forum where these approaches can be placed side by side with the EU AI Act and examined for compatibility rather than ranked for compliance. Comparison, not hierarchy, should be the operating principle.

The Dialogue's structure should reflect one basic insight: the people who build AI, the people who regulate it, and the people who live with its consequences rarely sit in the same room. When they do, the format usually ensures they talk past each other. Here is how to fix that. Member states should present not just national positions but specific governance problems they have failed to solve domestically. India, for example, is struggling with how to classify AI intermediaries under its IT Act. The EU is facing implementation challenges with the AI Act's high-risk classification. Honest problem-sharing is more useful than polished national statements. The private sector should be required to submit transparency disclosures as a condition of participation. Not full trade secrets, but basic information: what training data categories were used, what languages are supported, what safety evaluations were conducted. If companies want a seat at the governance table, they should bring evidence, not just talking points. Academia should be given a dedicated submission track for policy-relevant research, separate from civil society statements. Universities produce a different kind of input than advocacy organizations. Mixing them into the same "stakeholder" category flattens both contributions. A structured call for policy briefs from university-based researchers, with regional quotas to ensure Global South participation, would immediately improve the evidence base. For format, the Dialogue should avoid long plenary sessions with three-minute national statements. Instead, it should organize at least half the programme as thematic working sessions with mixed seating: one diplomat, one technologist, one academic, and one civil society representative per table, working through a specific governance problem with a defined output. The Geneva session should produce problem-specific working papers, not a single consensus document that says everything and commits to nothing.

Three groups are consistently missing from AI governance conversations, and their absence distorts the outcomes. First, legal academics and practitioners from developing countries. AI governance is fundamentally a legal design problem. It involves liability, intellectual property, data rights, and administrative law. But the lawyers and legal scholars shaping global AI norms are overwhelmingly based in a handful of jurisdictions. India has one of the world's largest legal professions and a rich tradition of constitutional rights jurisprudence, yet Indian legal scholars are rarely present in global AI governance forums. The same is true for legal academics in Brazil, Nigeria, Kenya, and Indonesia. The Dialogue should create a Legal Scholars Advisory Track with regional representation, specifically inviting law faculty who teach AI-related courses to contribute written inputs and participate in thematic sessions. Second, non-English-speaking communities affected by AI deployment. Large language models perform poorly in most Indian languages, African languages, and indigenous languages worldwide. Governance discussions about bias, fairness, and cultural impact are conducted almost entirely in English, by people whose languages are well-served by existing AI systems. The Dialogue should commission the Scientific Panel to produce a specific assessment of AI performance disparities across language groups and use this as a basis for discussion. Third, students and early-career researchers. The people who will implement whatever governance frameworks emerge from these discussions are currently in universities. They are absent from the conversation. At GD Goenka University, our AI Policy Lab involves law students directly in governance research. The Dialogue should establish a Youth Research Fellowship that funds 20 to 30 students from developing countries to attend the Geneva session and present their work. This is not symbolic inclusion. It is pipeline-building for the next generation of AI governance professionals.

Most international governance meetings follow a format designed in the 1960s: opening plenary, national statements, panel discussions, closing plenary. This format rewards prepared positions and discourages genuine exchange. The AI Dialogue should break from it in three specific ways. First, run governance simulations. Pick a real cross-border AI dispute, anonymize the parties, and ask mixed teams of diplomats, technologists, academics, and civil society representatives to negotiate a resolution in 90 minutes. For example: a generative AI model trained in Country A scrapes copyrighted works from Country B's creators and is deployed as a service in Country C. Who has jurisdiction? What remedy exists? This kind of exercise exposes governance gaps faster than any panel discussion and produces concrete problem statements that can feed into the Dialogue's output. Second, use a "red team" format for the Scientific Panel's report. Instead of a traditional presentation followed by questions, assign three or four teams drawn from different stakeholder groups to challenge specific findings in the Panel's inaugural report. One team argues the findings understate risk. Another argues they overstate it. A third tests whether the recommendations are implementable in a low-resource country. This turns passive reception of a report into active engagement with its substance. Third, create asynchronous policy labs that run for two months before the Geneva session. Assign each lab a specific governance question drawn from the Dialogue's themes. Participants contribute written inputs, respond to each other, and produce a short synthesis paper. By the time they arrive in Geneva, they have already worked through the substance. The in-person session becomes a place for decision-making, not first-contact brainstorming. The AI Dialogue should treat the Geneva meeting as the end point of a deliberative process, not the beginning.

Five examples stand out, not because they are perfect, but because they offer transferable lessons. India's AI Governance Guidelines (February 2026) adopted a "law-plus" model that layers voluntary governance on top of existing statutes rather than creating a standalone AI law. The practical insight here is that developing countries with limited regulatory capacity can govern AI without waiting for comprehensive legislation. The guidelines also proposed an India-specific risk classification grounded in social context rather than system type. This is a genuine alternative to the EU's approach and deserves serious comparative study at the Dialogue. Brazil's AI Bill (PL 2338/2023) is notable because it was shaped by extensive public consultation, including input from indigenous communities and labour unions. Most AI governance frameworks are written by technologists and lawyers. Brazil showed that broader participation does not slow down the process. It improves it. The OECD AI Policy Observatory remains the best comparative tool for tracking national AI governance approaches across 70 countries. The Dialogue should formally partner with the Observatory to avoid duplicating data collection and instead focus on what the OECD cannot do: convene the countries that are not OECD members. Japan's broad text and data mining exception under its 2018 Copyright Act amendment is the most permissive approach globally to AI training on copyrighted works. Whether one agrees with it or not, it has produced measurable effects on Japan's AI research output. The Dialogue should examine this as a case study in how intellectual property choices shape AI capacity. At a smaller scale, our AI Policy Lab at GD Goenka University demonstrates that governance research capacity can be built at zero institutional cost by embedding policy research into existing law courses. This model is replicable across hundreds of universities in the Global South without waiting for external funding.