CONSUMER UNITY AND TRUST SOCIETY (CUTS INTERNATIONAL)

Crisis management and geopolitical risk Crises are high-value precisely because they are rare, consequential, and genuinely novel. A crisis that closely resembles a previous one gets managed adequately by institutional memory and established protocols. The truly dangerous crises - the ones that destroy companies and governments - are the ones that don't resemble anything that came before, or that combine familiar elements in unfamiliar configurations. AI systems trained on historical crises will pattern-match to the closest precedent. Sometimes that is useful. But the most consequential error in crisis management is confidently applying the wrong historical analogy. Human crisis managers with deep domain experience do something different - they maintain uncertainty, they probe for what's different about this situation, they update rapidly on thin evidence that something is off. This is exactly the adaptive capacity that retraining cycles cannot replicate. A geopolitical realignment that rewrites supply chain logic overnight Supply chains are built on assumptions about political stability, trade relationships, and the reliability of particular routes and partners. Those assumptions get embedded in capital investment decisions, supplier contracts, and operational infrastructure over years and decades. When geopolitical realignment happens - a war, a sanctions regime, a diplomatic rupture, a sudden shift in trade policy - the assumptions embedded in existing supply chains can become invalid simultaneously and without warning. Companies that read the new configuration correctly and move fast capture an enormous advantage, and the companies that apply the old logic to the new situation suffer disproportionately because of their models, advisors, and institutional memory. Regulatory and Legal navigation High-stakes legal and regulatory work derives its value from operating at the frontier - in the gaps, ambiguities, and contested interpretations where rules run out. Outcomes there turn on creative argument, precedent synthesis, and relationship knowledge. That's precisely where established rules, and the AI trained on them, offer the least guidance.

• Safe, secure and trustworthy AI
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Interoperability of governance approaches
• Transparency, accountability, and human oversight

Under the proposed thematic clusters, 4(f), there is an urgent and pressing need for the transparency, accountability and robust human oversight of artificial intelligence systems in a manner that complies with international law, particularly in the Novel tail-risk AI processes that represent decisions made at scale, at speed, with systemic consequences, by systems whose behaviour is not fully understood even by their creators. In every other domain where humanity has faced decisions of this character - nuclear weapons, pandemic pathogens, climate-altering industrial processes the international community has eventually recognised that no single state or actor can be trusted to govern the risk alone, because the consequences of failure are borne collectively while the decisions are made unilaterally. The governments, irrespective of geography, have to be made to understand the importance of this. AI has reached that threshold. The novelty is real. The tail risks are real. The systemic interconnection is real. The concentration of decision-making in a handful of actors whose incentives are not aligned with collective welfare is real. International law exists precisely for this situation - to assert collective human authority over decisions whose consequences are collective. The framework is not just required. In the fullest sense, it is overdue. The specific risks that make international law utmost required in this segment are the irreversibility, systemic interconnection, opacity and verification difficulty, accountability vacuum, and the concentration of capability. An international law with a process-based rather than application-based approach, a mandatory pre-deployment review and a need to create a binding incident reporting obligation can address this. The global aviation safety model - where incident reporting is mandatory, protected, and feeds into global safety improvements - is the relevant precedent. Since such crises are cross-border and of an insurmountable proportion, it is beyond the control of one nation; international law is the only viable option in the hands of humanity.

Cross-cutting issues Human in the Loop shifted to Human Defining its linkage with Return on Investment (ROI). The bull case for sky-high tech valuations rests substantially on AI delivering transformative productivity gains and revenue expansion. But that outcome is not guaranteed - it is contingent on AI systems actually working reliably and at scale in real enterprise environments. And here is where "human in the loop" becomes financially material, not just philosophically interesting. This is more relevant in high novel tail-end domains where the commercial proposition is high. Innovation in AI is shifting to Impact, leading the scale. The AI value thesis is shifting from innovation to impact, and impact is where scale actually lives. LLMs have failed to deliver the scale their investment implied. SLMs, purpose-built for specific linguistic and contextual realities, are achieving it. India's Bhashini is the proof case: high reach, low cost, measurable ROI - the metrics LLM deployments have struggled to demonstrate. The implication for Big Tech is structural; the Global South is not a secondary market; it's the scale opportunity. A marriage between large foundation models and lightweight, locally-tuned systems - on Global South terms - may be the only credible path to the returns that current AI valuations require. With the AI bubble as backdrop, this isn't just strategic optionality. It's solvency arithmetic. The current gap between a P/E of 22 and 45 represents an enormous amount of market capitalisation resting on an assumption about AI's autonomous capability trajectory that has not yet been demonstrated and may be structurally more difficult than currently priced. Cross-border regulatory sandboxes, regulatory harmonisation and mutual recognition, skills programmes, South-South Cooperation in AI Policymaking: Developing a Collaboration Roadmap, Competing to Innovate: How Competition Accelerates AI Innovation, The AI-DPI Nexus: The Future of Public Interest Technology, Unlocking Impact: A Shared Data Platform for AI Innovation in Development Cooperation, World Café Cross-Border Applied AI Joint Research, innovation and startup Communities, turning expertise into practical next steps for international AI collaboration and digital Public Goods for Global AI Equity. Data sovereignty strategy operates simultaneously at the legislative layer, the infrastructure layer (sovereign compute capacity, data centre buildout), the diplomatic layer (bilateral chip access negotiations, adequacy frameworks), and the standard-setting layer (interoperability norms, open model architectures and is distributed across borders.

India has the DPDPA and an evolving AI governance framework. Still, the most serious accountability deficit is in government AI- welfare systems, policing tools, and Aadhaar-linked delivery, where the chain, although widespread from harmed citizen to meaningful remedy, is either absent or functionally unreachable. The Global South: compounded vulnerabilities Three factors make it worse beyond India. Foreign-built systems are opaque, and their developers untouchable by local law. The state, not the private sector, is the dominant AI deployer - and operates with even less transparency than firms do. And AI systems trained on Global North data carry biases that invisibly harm Global South populations, with no international standard requiring disclosure of demographic performance before deployment. Significant Challenges Lack of Human Oversight & Accountability: Transparency and Data Bias: AI systems, often trained on Western data, pose ethical risks, introduce bias, and raise privacy concerns in diverse societies. Fragmented Governance in India: While pursuing digital initiatives, India faces challenges from fragmented data, bureaucratic inefficiencies, and a need for better policy integration across states. Global South/India Constraints: Institutional capacity limitations, political/economic deterioration, and lack of staff skills for deploying and regulating AI are critical hurdles. Opportunities Digital Transformation: AI offers opportunities to optimise bureaucratic procedures, enhance public service delivery, and boost economic development, setting global norms: India is increasingly a leading player in defining global AI norms. Key Developments Localisation Gaps: There is a growing gap between international AI standards and their implementation, with local adaptation of AI policies remaining uneven. Regulatory Focus: Increased focus on creating national AI strategies that include accountability mechanisms, ensuring AI systems align with local social and legal contexts.

UN is still the apex platform for all communication and dialogues happening in all spheres of life. Countries across are not yet writing binding AI laws together -instead, they are talking. And right now, that talking is the most important thing happening in global AI governance. A series of international summits - Bletchley, Seoul, Paris, New Delhi — has gradually shifted the conversation from "how do we make AI safe" to "who gets a say in how AI is governed." India hosting the 2026 summit was itself a statement: the Global South will no longer accept rules written by others and handed down later. These dialogues have produced real results. Shared language like "sovereignty over data" and "accountability by default" now exists because countries talked until they found formulations everyone could accept. Ideas from outside the Western mainstream — India's open digital infrastructure model, African data governance frameworks — got onto the global agenda through these conversations, not through formal negotiations. But the limits are just as real. Dialogue produces declarations, not enforcement. The people most harmed by ungoverned AI — a farmer denied a subsidy, a worker scored by an algorithm — have no seat at these forums. And the companies actually building and controlling AI attend as guests, not as parties who can be held to account. The bottom line: these dialogues are necessary and useful, but they are not sufficient. They are slowly building the political ground for real enforceable rules — but while that ground is being built, the decisions that will shape AI for decades are being made by a handful of companies in two countries, with no meaningful global accountability over them.

Some of the existing initiatives, partnerships, or mechanisms that the AI Dialogue should build upon or connect with include: The AI Dialogue can leverage established global initiatives for greater impact. Building on these would amplify collaborative efforts in AI governance and ethics. Key Initiatives Partnership on AI (PAI): Multisector group with companies (e.g., Google, Microsoft), non-profits (e.g., ACLU, UNICEF), and experts fostering best practices, working groups on AI safety, and fellowships for civil society. Global Partnership on AI (GPAI): OECD-led network of 29+ countries advancing trustworthy AI through research, policy, and projects on ethics and societal impact. AI for Good (ITU): UN platform uniting stakeholders for responsible AI in human rights, poverty alleviation, and sustainable development via workshops and prototypes. Added Value The AI Dialogue could introduce fresh, inclusive forums bridging human rights experts, AI developers, and grassroots voices. It might add adaptive, AI-powered personalisation in dialogues for peace processes or education, enhancing scalability and real-time collaboration beyond static models. This would fill gaps in majority-world contexts and dynamic group simulations. Existing initiatives like Partnership on AI and GPAI provide strong foundations in multistakeholder ethics and policy, but often prioritise Western tech giants, risking echo chambers and limited Global South input. Critically, they lack agile, real-time mechanisms for crisis-driven dialogues (e.g., peace processes) and grassroots scalability.

Effective AI dialogue requires moving beyond passive consultation toward active, iterative co-design. The primary challenge in current dialogues is the "asymmetry of influence": technical stakeholders often dominate the conversation, while civil society and the public are frequently brought in too late, often only to react to finished products. Critical Examination of Stakeholder Contributions To be effective, each stakeholder must move beyond their traditional siloed roles. Currently, many contributions are reactive or performative. The following table highlights the ideal contribution versus the common pitfall: Stakeholder Ideal Contribution Common Pitfall Developers/Tech: Provide technical reality-checks, safety parameters, and scalability insights."Technological solutionism"—assuming code can solve complex, non-technical social problems. Policymakers create flexible, future-proof regulatory guardrails and enforcement mechanisms. Regulatory capture or over-regulation that stifles beneficial innovation due to lack of technical literacy. Academia/Civil Society Stress-test models for bias, rights, and long-term societal impact. Being relegated to "ethics washing" where they are included for optics but ignored in practice. General Public Provide "lived experience" feedback regarding utility, accessibility, and trust. Being treated as passive consumers rather than active participants with agency. The format must be Transition to "Iterative Deliberation", Implement "Red-Teaming" Public Consultations, Embed Interdisciplinary "Ethics Liaisons" and Incentivise Diverse Participation.

Underrepresented voices in AI governance in descending order are: Informal economy workers - the largest affected group globally, with no representation mechanism and no way to translate lived experience into policy. Rural and subsistence communities - smallholder farmers and village populations whose access to land, credit, and welfare is increasingly AI-mediated, yet who are unreachable by every forum where rules are made. Indigenous peoples - doubly excluded: their data is extracted without consent, and their governance traditions are invisible in frameworks built around nation-state sovereignty. Women in low-income contexts are disproportionately harmed by biased AI in credit, health, and justice systems, and are underrepresented in both the technical workforce and policy rooms. Displaced persons and stateless populations - among the most heavily surveilled and AI-processed groups, yet with no legal standing to challenge the systems applied to them. Linguistic minorities - AI performs worst in their languages, and the governance documents shaping their AI environment are written in languages that most cannot read. People with disabilities are excluded from both the design of AI systems that govern their access to services and the forums that set the standards for those systems. Youth in the Global South - the generation that will live longest with today's decisions has the least institutional access to shape them. Small Global South governments - fully subject to the international AI order but too small and under-resourced to meaningfully influence it. Civil society and academia outside the North Atlantic - not for lack of expertise, but for lack of funding, visa access, and institutional recognition in global standard-setting processes. The common thread: those most exposed to AI's consequences are least present when its rules are written.

Moving beyond standard meetings, innovative engagement in 2026 leverages technology-enabled deliberation and immersive participation to bridge the gap between complex technical development and public trust. Innovative Engagement Formats for AI Dialogue must include: Citizens' AI Assemblies, "Red-Teaming" Hackathons, AI Sandboxes (Participatory), Collective Sentiment Mapping, Virtual Reality (VR) Simulations, Algorithmic Auditing Sprints, Core Principles for Dynamic Engagement Asynchronous Accessibility: Use digital portals (like the UN's 2026 Global Dialogue portal) to allow global participation across time zones. Structured Feedback Loops: Dialogue is only "dynamic" if participants see their input reflected in the next version of the model or policy. Gamification of Ethics: Using "serious games" to help non-experts understand the trade-offs in AI training (e.g., Privacy vs. Accuracy).

Newer approaches advancing AI governance - including last-mile populations 1.India's DPI model - open, interoperable public infrastructure (UPI, Aadhaar, ONDC) built to prevent platform capture and keep services accessible to those outside the formal economy. Now proposed as a global template. 2.Subsidised compute access--India AI's ₹65/hr GPU pricing democratises AI development beyond hyperscaler reach. The Delhi Summit's proposed Global Compute Bank extends this internationally, treating compute as a development resource. 3.Multilingual model development - BharatGen across India's 22 languages, Masakhane in Africa, SEA-LION in Southeast Asia -directly addressing the exclusion of non-English-speaking communities from AI's benefits. 4.CARE Principles - an indigenous-led data governance framework asserting community sovereignty over data. Now referenced in national strategies across New Zealand, Canada, and Latin America. 5.Africa's AU Data Policy Framework - a continent-wide governance baseline enabling smaller African nations to adopt shared standards without building costly bespoke frameworks, creating a South-South adequacy pathway. 6.Feminist AI initiatives- UNESCO's AI Ethics Recommendation, the Feminist AI Research Network, and the AI Gender Equality Index building the evidence base for gender-responsive AI procurement. 7.Algorithmic impact assessments - Canada's mandatory disclosure model for government AI, now being adapted by Rwanda, Uruguay, and Estonia for their own public sector contexts. 8.Community-based AI auditing - grassroots initiatives training local organisations to identify and document algorithmic harm, building accountability from below rather than waiting for regulatory capacity from above. 9.EU AI Act as a reference floor - its extraterritorial reach gives Global South civil society a standard to cite domestically, with the Brussels Effect operating through advocacy even where law doesn't apply. 10.Disability-inclusive design standards -Partnership on AI and Global Disability Innovation Hub frameworks treating accessibility as a baseline requirement, not an afterthought. 11.Offline-first AI tools - designed for low-end devices, low connectivity, and local languages, with Google's Project Euphonia and Microsoft's Rural AI programme as early examples of infrastructure meeting governance intent.