SFLC.in

(i) Ensuring Global South has an equal seat at the table: Discussion on Global South is often part of global conversations but Global South Actors participation continues to suffer. AI governance discussions are still heavily dominated by Global North actors. The UN can help correct this imbalance by institutionalizing Global South representation in all major AI governance dialogues, standard-setting processes, and multistakeholder consultations. Asian markets must be understood and studied separately. (ii) Facilitate meaningful engagement between civil societies, Company and governments - UN can use their credibility and neutrality to host national level events and create a space for open communications to discuss human rights risks in AI advancements. Also convene dialogues between gig worker unions, labour ministries, and tech companies to address labour issues. Also convene dialogues between gig worker unions, labour ministries, and tech companies to address labour issues specific to AI. (iii) Address data extraction, linguistic injustice, and the push of "free" AI tools in Asia - There is growing concern that free AI tools are being introduced to collect vast amounts of linguistic and behavioural data especially in markets with weaker data protection standards. UN can help interrogate these practices and encourage companies to invest meaningfully in indigenous and low-resource language datasets and content moderation rather than relying on shortcuts. (vi) Creation of a Model AI Law: Countries are drafting AI laws and amending copyright statutes independently, leading to fragmentation. There is a clear role for the UN to support the development of a model AI law and provide guidance on copyright reform, so that states have a foundation to build on. This could be similar to the UNCITRAL Model Law on Electronic Commerce (1996) which paved the way for Internet regulations in many countries.

• Open-source software, open data and open AI models
• Transparency, accountability, and human oversight
• Protection and promotion of human rights
• Social, economic, ethical, cultural, linguistic and technical implications of AI

1. Protection and promotion of human rights AI governance must remain firmly grounded in human rights frameworks, particularly the UN Guiding Principles on Business and Human Rights, which Member States have already endorsed. As AI systems scale globally, they are already producing visible harms from labour exploitation in data supply chains to online gender-based violence and systemic discrimination. Ensuring dignity, safety, and access to remedy especially for invisible workers such as content moderators and data annotators in the Global South must be central to governance efforts. 2. Transparency, accountability, and human oversight A key challenge in AI governance is the opacity of both systems and the underlying labour and data infrastructures that sustain them. There is an urgent need for enforceable transparency across AI supply chains, regular disclosure of AI-related incidents, and independent auditing mechanisms. Human oversight must remain non-negotiable, particularly in high-risk systems, alongside meaningful grievance redress mechanisms and post-deployment monitoring. 3. Social, economic, ethical, cultural, linguistic and technical implications of AI AI is not merely a technical system but one embedded in social and economic realities. In Asia and other Global Majority contexts, this includes job displacement, exploitative labour practices, environmental strain, and linguistic marginalisation due to the dominance of high-resource languages. The deployment of "free" AI tools also raises concerns about data extraction and inequitable value flows. Addressing these structural and contextual impacts is critical to ensuring that AI does not deepen existing inequalities. 4. Open-source software, open data and open AI models Open AI ecosystems can play a key role in addressing concentration of power and enabling more equitable participation in AI development. Open models and datasets support transparency, independent auditing, and locally grounded innovation particularly for low-resource languages and contexts. However, openness must be accompanied by safeguards to ensure accountability and responsible use.

1.The political economy of AI cuts across all themes but is not directly named. Questions of market concentration, infrastructural dependency, and unequal bargaining power between Global North firms and Global Majority countries shape how AI is developed and governed. Without addressing these structural imbalances, efforts on transparency or human rights risk being limited in practice. 2.Invisibility of labour in AI systems requires sharper focus. While labour concerns sit within broader social and economic implications, the scale and severity of harms in AI supply chains from content moderation to data annotation suggest the need for more explicit recognition. These systems depend on outsourced, precarious workforces that remain largely excluded from governance conversations, with limited protections or avenues for remedy. 3. Data governance gap, particularly in emerging markets, is insufficiently foregrounded. The growing deployment of "free" or subsidised AI tools raises concerns around large-scale data extraction, weak consent frameworks, and the use of diverse linguistic and behavioural data to train global models without equitable benefit-sharing. 4.There is a need to better articulate the limits of techno-solutionism in AI governance. Current approaches often over-rely on technical fixes such as automated detection or auditing tools-while underestimating the role of legal, institutional, and democratic oversight. Governance must remain a socio-technical exercise, not one delegated primarily to private actors or technical systems. 5.Access to remedy and grievance mechanisms remains underdeveloped across frameworks. Even where harms are recognised, affected individuals and communities particularly in the Global South often lack clear, accessible pathways to challenge or seek redress.

(i) On Labour: Job displacement issues are a major concern in Asia as the job market thrived on outsourced low skill tech jobs which are now being automated. Along with it, the labour exploitation issue is prevalent in Asia. The exploitation of the tech industry continues to exacerbate with AI Development: For global companies, Asia represents a large pool of inexpensive, compliant labor. AI-related labour practices in Asia raise serious human rights concerns. Millions of content moderators and data annotators especially in India and the Philippines work under harsh conditions that violate basic rights to fair wages, safe work, and dignity. These workers perform essential tasks such as reviewing traumatic content and labeling data to train AI systems, yet they receive very low pay, have no job security or social protections, big techs hire them through opaque outsourcing chains that keep them invisible and unprotected. Many face severe mental health impacts from constant exposure to violent or abusive material without adequate psychological support. When harm occurs, there are almost no grievance mechanisms or avenues for effective remedy. This reliance on cheap, disposable labour in the Global South reflects structural inequalities and allows global companies to avoid accountability, especially in countries where labour regulations are weak. (ii) Environmental concerns for Asia - Asia has the world's largest population and already struggles with limited resources. As AI expands, it puts even more pressure on resources. Without proper planning, the energy demands of AI could worsen sustainability challenges, raise operational costs, and undermine climate commitments across the region. (iii) Multilingual and Low-Resource Language Challenges: Automated moderation systems used by Platforms are trained on high-resource languages like English, or widely spoken languages like European languages, indigenous languages lack sufficient training data and companies lack financial incentives to invest in moderation resources for less profitable markets in the Global South. This technological gap results in slower moderation responses, higher error rates, censorship or shadow banning of minority language speakers. It also exacerbates political conflicts, misinformation and surpasses marginalized voices. (iv) Misuse of Asian markets for AI model Training: Companies like OpenAI, Google, and Perplexity are providing premium AI tools for free or subsidized prices in the Indian market. This is similar to Facebook's free basics initiative which offered free access to certain websites and was objected in various countries including India as violative of Net Neutrality principles The free access to AI technologies raises questions about the underlying incentives for big techs - whether countries like India are being targeted as a place to gather diverse data, refine models, and test AI use cases that could later scale across other emerging markets especially in regions with absence of comprehensive data protection laws. This also raises important questions about data privacy and the ethical use of personal information.

The UN is uniquely positioned, given its credibility and neutrality, to convene governments, companies, civil society, and labour representatives in meaningful dialogue. This includes creating national and regional platforms to openly discuss human rights risks in AI systems, as well as facilitating focused engagements between gig worker unions, labour ministries, and technology companies to address labour issues emerging from AI-driven systems. The Dialogue can institutionalise Global South participation across standard-setting processes, multistakeholder consultations, and governance forums. This must go beyond symbolic inclusion to ensure agenda-setting power, with particular attention to regions like Asia, whose markets and regulatory approaches require independent and context-specific engagement. The Dialogue can facilitate the development of shared resources—such as a model AI law and guiding principles on copyright—providing states with a common foundation while allowing for contextual adaptation. There is already broad, if uneven, commitment to principles such as human rights, safety, and accountability. The Dialogue can translate these into more operational guidance—clarifying what meaningful transparency, human oversight, or due diligence should look like across jurisdictions—while allowing states to adapt implementation to local contexts.The Dialogue can support alignment on standards, audit practices, and risk classifications, reducing fragmentation without forcing convergence.

A key foundation is the UN Guiding Principles on Business and Human Rights, which provide a widely endorsed framework for corporate responsibility and state duty. As companies increasingly align their AI principles with the UNGPs, the Dialogue can help translate these commitments into more concrete expectations for AI systems particularly around due diligence, human rights impact assessments, and access to remedy. Civil society and labour-led initiatives have also been critical in documenting harms across AI supply chains, particularly in content moderation and data annotation, though these perspectives are often not meaningfully integrated into formal governance spaces. The added value of the AI Dialogue lies in its ability to connect these parallel efforts. It can act as a bridge between normative frameworks (like the UNGPs), technical initiatives, and lived realities—particularly in the Global South. Most importantly, the Dialogue can move beyond voluntary commitments by encouraging greater coherence, accountability, and interoperability.

A useful way to structure the AI Dialogue is to centre it on co-drafting and negotiated outcomes, rather than sequential consultations. The cohort can be divided into three parallel working groups, each tasked with developing a Model AI Law. Crucially, each group should have equal representation from government, civil society, and industry, ensuring that diverse perspectives are embedded within each drafting process rather than introduced later. This design encourages early negotiation on difficult issues—such as liability, data governance, labour protections, and transparency—within each group. Day 1: Co-drafting phase Each group operates as an independent drafting body and produces its own version of a Model AI Law. The aim is not uniformity, but to surface different approaches that are nonetheless internally balanced across stakeholder interests. Governments contribute regulatory feasibility, industry brings technical and operational insight, and civil society anchors the framework in human rights and public interest considerations. Day 2: Negotiation and convergence The three groups then enter a structured negotiation phase. Here, they present their respective frameworks, identify points of convergence and divergence, and work toward a consolidated model law. This process should be facilitated to ensure that disagreements—particularly on politically or economically sensitive issues—are meaningfully addressed rather than deferred. The outcome would ideally include: A joint model AI law reflecting negotiated consensus; A record of contested issues and alternative approaches, where agreement is not possible; Clear indications of implementation pathways adaptable across jurisdictions. This format ensures that the Dialogue produces tangible outputs while also modelling the kind of multistakeholder negotiation that effective AI governance will require in practice.

First, workers across AI supply chains—including content moderators, data annotators, and gig workers—are largely absent, despite being essential to how AI systems are built and maintained. Their working conditions, mental health impacts, and lack of labour protections rarely shape governance frameworks. Including them requires direct representation through unions or worker collectives, as well as structured consultations that go beyond anonymised or extractive research. Second, Global South actors, particularly from Asia and Africa, remain underrepresented in agenda-setting spaces. While often discussed, they are seldom positioned as decision-makers. Inclusion here must move beyond participation toward institutionalised representation in standard-setting bodies, regional consultations, and leadership roles within global forums. Regional diversity—especially linguistic and socio-economic differences within Asia—must also be acknowledged. Third, linguistic and culturally marginalised communities are frequently overlooked. AI systems continue to privilege high-resource languages, leading to uneven moderation, exclusion, and misrepresentation. Including these communities requires investment in local language expertise, community-led datasets, and participatory approaches to system design and evaluation. Fourth, civil society organisations working on intersecting harms—such as gendered disinformation, disability rights, and environmental impacts—often lack sustained access to governance spaces. Their inclusion should be supported through funding, capacity-building, and formal roles in multistakeholder processes. Finally, there is limited representation from countries with low regulatory and technical capacity, which face the dual challenge of rapid AI adoption and weak institutional safeguards. Targeted capacity-building, peer learning platforms, and equitable access to governance tools are essential.

Co-drafting lab, where participants are placed into mixed stakeholder groups (government, civil society, industry) and tasked with producing concrete outputs—such as a model AI law or regulatory framework. This shifts engagement from abstract principles to negotiated trade-offs, compelling participants to reconcile technical feasibility, rights protections, and policy constraints in real time. Regional deep-dive roundtables are essential to ground global discussions in local realities. These should be designed as closed, trust-based spaces where participants can speak candidly about context-specific risks such as labour exploitation, linguistic marginalisation, or data extraction without the pressure of formal positioning. Scenario-based simulations -Participants can be given realistic case studies—such as a harmful AI deployment, a data breach, or algorithmic bias in a public system—and asked to respond as regulators, companies, and affected communities. This helps surface gaps in existing governance approaches, particularly around accountability, crisis response, and access to re

A foundational framework is the UN Guiding Principles on Business and Human Rights, which increasingly inform how companies approach AI-related risks. Their emphasis on human rights due diligence, impact assessments, and access to remedy provides a practical structure for embedding accountability into AI development and deployment. At the regulatory level, the EU AI Act demonstrates how a risk-based approach can be operationalised-linking obligations such as transparency, documentation, and human oversight to the level of risk posed by specific AI systems. While not without challenges, it offers a concrete template for aligning innovation with safeguards. On the technical and institutional side, participatory auditing models such as ParakhAI in India point toward more context-sensitive governance tools. By combining risk assessment (bias, privacy, security) with multistakeholder input and local language capabilities, such platforms help bridge the gap between high-level regulation and real-world system evaluation. There are also emerging transparency and openness practices. Industry-led efforts-such as publishing model cards, safety evaluations, or releasing open-source models-can enable independent scrutiny and broader participation. While uneven, these practices can support accountability when paired with clear standards and external oversight.