Global Artificial Intelligence Accountability Law and Governance Institute

The first Global Dialogue on AI Governance will be a success only if it produces binding commitments rather than aspirational declarations. The credibility of this forum, the first universal platform where every nation has an equal seat at the table, depends entirely on whether it produces obligations, not only aspirations. Concrete success outcomes must include: first, a formal roadmap toward a binding international AI governance instrument establishing minimum accountability standards, particularly for high-risk and agentic AI systems. Second, agreement on a common doctrine of graduated systemic liability for AI-caused harm, distributing accountability across developers, operators, deployers, and users in proportion to foreseeability, control, and benefit. Third, the establishment of an International AI Dispute Resolution Mechanism for cross-border AI harm claims. Fourth, a UN-led AI Law and Governance Capacity Programme specifically for developing nations, with a focus on training AI law practitioners, regulatory staff, and judicial officers, addressing the governance capacity gap, not merely the technical one. Fifth, adoption of mandatory provenance marking and disclosure standards for AI-generated synthetic media. Sixth, an international AI rights of redress framework granting individuals legal standing to challenge AI-mediated decisions affecting fundamental rights, regardless of where the system was developed. The pace of AI capability development is already outrunning the pace of governance development, and that gap is widening. If the Dialogue adjourns with voluntary guidelines and a commitment to further consultation, it will have failed the billions of people, disproportionately in the Global South,who bear the greatest governance risks and have the least institutional recourse.

• Safe, secure and trustworthy AI
• AI capacity-building
• Protection and promotion of human rights
• Transparency, accountability, and human oversight

Safe, Secure and Trustworthy AI This is the bedrock, but current governance frameworks contain a critical blind spot: agentic AI systems capable of taking sequential autonomous decisions and causing harm without direct human instruction at each step. Existing liability frameworks , whether in tort, product liability, or contract, were designed for a world in which human agency is identifiable at the point of harmful action. Agentic AI breaks that assumption entirely. Safety governance must evolve accordingly. Protection and Promotion of Human Rights AI systems trained on non-representative data, deployed without local legal accountability, and governed by foreign regulatory frameworks produce outcomes that entrench historical inequalities. Across the Global South, affected individuals have no practical recourse when AI systems deny them credit, flag them incorrectly for law enforcement, or exclude them from public services. Human rights frameworks designed for human actors must be expressly extended to AI-mediated decisions. Transparency, Accountability, and Human Oversight Transparency without enforceability is not accountability. Moreover, human oversight that is structurally nominal, where the human in the loop lacks technical capacity to evaluate or override the system, is not a governance safeguard. It is liability transfer. International standards must include minimum competency and procedural requirements, not merely the formal presence of a human decision-maker. AI Capacity-Building The governance capacity gap is as urgent as the technical one. Most developing nations lack trained AI law practitioners, AI-literate judicial officers, and AI-competent regulatory bodies. India's Digital Personal Data Protection Act, 2023, illustrates the challenge: sound legislation exists, but effective implementation requires a class of practitioners that does not yet exist in sufficient numbers. This is the norm across the Global South.

Several critical issues remain inadequately addressed: Agentic AI and the Accountability Void The most significant under-addressed gap is legal accountability for agentic AI systems. When an autonomous AI agent causes financial harm, discriminatory outcomes, or physical injury, accountability is diffuse across the developer of the foundation model, the operator who deployed it, the user who configured it, and the enterprise on whose behalf it acted. No existing legal framework assigns responsibility across this chain in a manner that is both just and enforceable across jurisdictions. A doctrine of graduated systemic liability, drawing on environmental and pharmaceutical law, is urgently needed. AI and Structural Cultural Harm AI is not culturally neutral. Models governing access to credit, healthcare, employment, and judicial processes were designed within the cultural, linguistic, and legal frameworks of a small number of high-income nations. When deployed across the Global South, they carry embedded assumptions that do not translate accurately, producing structural disadvantage encoded at scale. The erosion of linguistic diversity and indigenous knowledge systems constitutes a form of AI-mediated cultural harm that current governance frameworks do not recognise. Open-Source AI Accountability Gap Open-source AI models are presented as solutions to access inequality, but open weights without open accountability create a specific governance problem. The ability to fine-tune and deploy a powerful model carries no obligation to ensure safe or legally compliant deployment. A model released under an open licence and fine-tuned for harmful application in a jurisdiction with no AI law creates a liability vacuum that urgently requires international standards addressing minimum documentation and risk-disclosure obligations. Deepfake and Synthetic Media Governance Existing defamation, copyright, and electoral interference laws were not designed for synthetic content generated at near-zero cost at industrial scale. The International AI Accountability Forum has advocated for a three-layer framework: mandatory provenance marking at generation, platform liability for amplifying unmarked synthetic content, and express criminal liability for synthetic content deployed to defraud, defame, or interfere with electoral processes. This framework warrants international adoption.

India, as the world's most populous nation and one of the fastest-growing digital economies, exemplifies both the transformative potential and the governance failures of AI deployment at scale. The most consequential governance gap is not technical; it is legal. India currently lacks a comprehensive, standalone AI governance law. The Digital Personal Data Protection Act, 2023, represents a significant step toward sovereign data governance, but its effective implementation requires a class of AI law practitioners, AI-literate judicial officers, and AI-competent regulators that does not yet exist in sufficient numbers. Legislation without enforcement capacity is not governance - it is aspiration. AI systems deployed across India in financial services, law enforcement, and public administration were overwhelmingly designed within non-Indian cultural, linguistic, and legal frameworks. The result is structural disadvantage encoded at scale: credit denial algorithms that do not recognise informal income patterns; facial recognition systems with elevated error rates for darker skin tones; judicial risk assessment tools trained on data that does not reflect Indian social realities. Across the broader Asia-Pacific and Global South region, the governance capacity gap is even more acute. Most nations lack the specialized regulatory expertise to draft, enact, or enforce AI legislation. AI governance remains, in practice, a domain governed by the regulatory choices of a small number of technologically advanced nations and a handful of dominant corporations, neither of which is accountable to the populations most affected. The opportunity is equally significant. India's scale means that governance innovations, in AI-specific liability frameworks, synthetic media regulation, and legal standing for AI-affected individuals, can be deployed at a scope that generates global learning. The Dialogue must create mechanisms to capture and disseminate that learning.

The AI Dialogue has a singular opportunity to establish the international cooperation architecture that voluntary guidelines and soft-law instruments have demonstrably failed to deliver. Its primary contribution must be to initiate negotiations toward a binding international AI governance instrument, one that establishes a floor of non-negotiable obligations while permitting regulatory variation above that floor. The deeper interoperability problem in AI governance is legal, not merely technical. Different jurisdictions assign liability differently, define prohibited AI applications differently, and impose disclosure obligations with different thresholds. A multinational enterprise deploying AI across jurisdictions currently faces genuinely incompatible legal obligations; the result, in practice, is regulatory arbitrage, where deployment is structured to minimize obligations rather than maximize safety. Only a binding international framework can address this systematically. The Dialogue can also establish an International AI Dispute Resolution Mechanism - a dedicated forum to adjudicate cross-border AI harm claims where national jurisdictions conflict or are absent. This would be a genuinely novel contribution that no existing mechanism provides. Critically, the Dialogue's cooperation architecture must structurally embed the Global South, not merely invite it as an afterthought. The window for establishing an international AI governance instrument on terms that genuinely reflect the interests of all nations is narrowing as AI capability development accelerates. The Dialogue must act with appropriate urgency , producing a concrete negotiating roadmap, not another statement of principles.

Several important foundations deserve explicit engagement: Council of Europe Framework Convention on AI (2024) The first binding international treaty on AI, establishing human rights, democracy, and the rule of law as non-negotiable foundations, provides the closest existing template for a UN-level binding instrument. The Dialogue should build explicitly on its accountability architecture and binding floor-obligation model. UNESCO Recommendation on the Ethics of AI (2021) The first global normative framework on AI ethics. The Dialogue's value-add is operationalizing its principles through enforceable obligations, particularly on cultural harm and indigenous knowledge protection. OECD AI Principles and G7 Hiroshima AI Process Provide governance frameworks among advanced economies. The Dialogue can universalize their most effective elements while correcting their exclusion of developing-nation priorities and their reliance on voluntary compliance. African Union Continental AI Strategy and ASEAN AI Governance Framework Regional approaches that must be integrated, not overridden, reflecting the reality that AI governance must accommodate different legal traditions including common law, civil law, and customary law systems. ISO/IEC JTC 1/SC 42 and ITU AI for Good Platform Technical standards infrastructure that legal frameworks must align with, particularly on provenance marking for synthetic media and risk classification thresholds for open-source models. The Dialogue's irreplaceable added value is delivering what none of these initiatives can: a universally applicable, legally binding, enforceable framework with equal standing for all nations, and institutional infrastructure, including a dispute resolution mechanism, to give it effect.

Effective AI governance requires genuine multi-stakeholder participation with structured accountability, not tokenistic consultation. Governments must participate at ministerial level with clear negotiating mandates, particularly toward a binding instrument. Civil society must hold formal, not merely observer, status, particularly organizations representing affected communities in the Global South. Academia and technical experts from developing nations must be structurally integrated, not simply invited when convenient. The private sector must participate under conflict-of-interest safeguards; AI developers and deployers are essential participants but must not function as agenda-setters. Legal practitioners and human rights defenders, especially those working at the intersection of AI accountability and access to justice in developing nations, represent a critically under-utilized voice. Structurally, the Dialogue should adopt a tiered model: a high-level intergovernmental negotiating track working toward binding commitments; a multi-stakeholder advisory council with formal input rights and published responses; thematic working groups producing actionable policy recommendations, including draft treaty language; and a global public participation platform with multilingual access. The format must combine in-person plenary sessions in Geneva with regional preparatory consultations ,in Africa, Asia-Pacific, Latin America, and the Arab States , to ensure geographic equity and translate global frameworks into regional realities. Asynchronous digital deliberation platforms must carry formal status, not merely informational value, to ensure that participation is not limited to those who can afford international travel.

The most critically underrepresented voices include: Communities in the Global South Subject to AI systems designed without their input, in languages they do not speak, reflecting values and institutional assumptions that do not represent them. The structural disadvantage this produces is the predictable result of governance frameworks designed by and for technologically advanced economies. Indigenous Communities Whose cultural heritage, traditional knowledge systems, and linguistic diversity are increasingly at risk from AI-powered extraction, surveillance, and the homogenizing effect of large language models trained on non-representative data. Free, prior, and informed consent principles must be expressly extended to AI deployment in indigenous territories. AI-Affected Individuals Without Legal Recourse The billions of people across the Global South who have been denied credit, flagged by law enforcement, or excluded from public services by AI systems against which they have no practical legal standing. Their experiences must directly inform governance frameworks. Legal Practitioners from Developing Nations Including AI law practitioners, judicial officers, and regulatory staff who understand governance gaps from the implementation side but are rarely present in technical AI forums dominated by engineers and policy specialists from high-income countries. Youth and Future Generations Whose lives will be most shaped by today's governance decisions. They must have dedicated participation mechanisms and formal input rights, not merely youth observer slots. Inclusion requires funded participation, language accessibility, guaranteed decision-making influence, and representation in the drafting of any binding instrument from the outset, not at the review stage.

AI Governance Liability Simulations Structured exercises in which stakeholders from different legal traditions navigate real agentic AI harm scenarios, mapping the chain of accountability across developer, operator, user, and enterprise, to build shared understanding of the graduated systemic liability doctrine and reveal gaps in proposed frameworks before they are adopted. Citizen Juries on AI Rights of Redress Randomly selected panels of ordinary citizens from diverse regions deliberating on what meaningful legal standing for AI-affected individuals should look like. This model, proven in climate and health governance, democratizes participation and surfaces lived-experience perspectives that expert consultations miss. Open-Source Accountability Challenge Structured expert sessions in which AI developers, lawyers, and civil society organizations collectively draft minimum documentation and risk-disclosure standards for open-source models above defined capability thresholds, thereby producing concrete treaty-ready language, not just recommendations. Deepfake Governance Workshop A dedicated format bringing together legal practitioners, platform representatives, electoral commissions, and affected individuals to stress-test the three-layer accountability framework, being provenance marking, platform liability and criminal liability, across different jurisdictions. Asynchronous Multilingual Deliberation Platform An accessible online forum enabling structured written input from any individual or organization globally, with formal mechanisms to incorporate outputs into the official negotiating process, thereby ensuring that participation is not rationed by geography or resources.

Council of Europe Framework Convention on AI (2024) The most important reference point -the first legally binding international treaty on AI, establishing that human rights, democracy, and the rule of law are non-negotiable foundations. Its structure of binding floor obligations with permitted national variation above that floor is the appropriate model for a UN instrument. EU AI Act (2024) Demonstrates comprehensive risk-based regulation with prohibited practices, mandatory transparency, human oversight requirements, and enforcement mechanisms. Its treatment of high-risk AI systems and general-purpose AI models is particularly instructive, though its design reflects European legal culture and cannot be transplanted wholesale to other jurisdictions. Indian Digital Personal Data Protection Act, 2023 Illustrates both the potential and the limits of national AI-adjacent legislation in the Global South. It represents genuine progress toward sovereign data governance but demonstrates that sound legislation requires an ecosystem of trained practitioners to be effective, a gap the Dialogue's capacity-building programme must address. Singapore's Model AI Governance Framework Offers a sector-neutral, practical approach widely adopted across the Asia-Pacific region, demonstrating that governance need not be burdensome to be effective. International AI Accountability Forum - Three-Layer Deepfake Framework Mandatory provenance marking at generation, platform liability for amplifying non-compliant content, and express criminal liability for synthetic content deployed to defraud or interfere with elections. A concrete, actionable model ready for international adoption. Rwanda's National AI Policy and Brazil's AI Bill Demonstrate that effective, rights-respecting AI governance frameworks are being developed across the Global South and must be integrated into , not merely acknowledged by, the global framework. The common thread across all effective approaches is the integration of legal enforceability, human rights centrality, technical standards alignment, and inclusive stakeholder participation, precisely the combination that only a binding international instrument, developed through a genuinely inclusive process, can deliver at scale. Dr. Pavan Duggal Founder President, Global Artificial Intelligence Accountability Law and Governance Institute Chair, International AI Accountability Forum Advocate, Supreme Court of India President, Cyberlaws.Net New Delhi, India | 30 April 2026