Independent Researcher — AI Ethics, Governance and International Human Rights Law

The first Global Dialogue will be a success only if it produces binding recommendations, clear timelines, and named accountability mechanisms — not a summary of shared concerns and a commitment to further dialogue. Governance has never kept pace with technology. We created the internet before we governed disinformation. We built social media before we protected children. We deployed algorithmic decision-making before we established due process. In each case, the argument was the same: the technology is moving too fast, voluntary measures will suffice, binding regulation can wait. In each case, the argument was wrong. Success requires four concrete outcomes. First, a formal finding that AI systems designed to maximise engagement at the cost of user mental health and epistemic autonomy are not safe or trustworthy — and that governing AI in the public interest requires prohibiting this design logic through binding standards, not voluntary commitment. Second, a commitment to comparative analysis of non-Western philosophical and legal traditions — including Vedantic, Buddhist, Ubuntu, and Indigenous frameworks — before any AI legal status or personhood framework proceeds internationally. Third, clear guidance from the Office of the High Commissioner for Human Rights and the ICC Prosecutor on how existing international law applies to AI-enabled violations in civilian contexts, requiring no new treaties — only the political will to apply law that already exists. Fourth, a formal acknowledgement that human rights obligations apply throughout the AI lifecycle beginning at training, not only at deployment — with binding labour standards for training workers as a first step. The test of this Dialogue is not whether governments had a productive conversation. It is whether the people paying the price for ungoverned AI are better protected after Geneva than before it.

• Safe, secure and trustworthy AI
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Protection and promotion of human rights
• Transparency, accountability, and human oversight

These four themes reflect the foundational gaps that, if unaddressed, will render all other governance efforts inadequate. Safe, secure and trustworthy AI is urgent because systems currently deployed are demonstrably unsafe for human wellbeing. AI-powered recommendation systems are optimised for engagement - deliberately exploiting documented psychological vulnerabilities including variable reward schedules and social validation loops, at measurable expense to users' mental health, epistemic autonomy, and democratic participation. This is not a byproduct of AI. It is an intentional commercial design choice. A system engineered to maximise the time a person spends on a platform, regardless of the harm that time causes, is not a trustworthy system. Binding design standards prohibiting engagement-maximising AI, particularly for users under 18, are the minimum condition for this commitment to have meaning. Social, economic, ethical, cultural, linguistic and technical implications of AI is the most systematically underaddressed theme in current governance discourse. The vocabulary of AI governance is not culturally neutral - it reflects specific Western philosophical traditions. The majority of the world's people live within frameworks - Advaita Vedanta, Ubuntu, Tibetan Buddhist philosophy, Indigenous legal traditions - that understand consciousness, personhood, and ethical responsibility differently. Excluding these from the governance conversation is a structural bias with direct consequences for who AI systems serve and who they harm. Protection and promotion of human rights is urgent because violations are already occurring in civilian contexts - algorithmic discrimination, mass surveillance of minorities, AI-assisted determinations denying refugee status - and not one is currently subject to effective international accountability. Existing law already prohibits this conduct. What is absent is enforcement. Transparency, accountability and human oversight are essential because voluntary commitments have failed. Binding instruments with independent audit requirements and effective penalty mechanisms are not a preference - they are the minimum condition for governance that is not complicit in the harms it claims to address.

The most significant cross-cutting issue not captured by the listed themes is the question of AI legal status and personhood - and the risk that personhood frameworks will be adopted internationally without adequate philosophical, cultural, or legal analysis of their consequences. Discussions of AI legal status are accelerating within legal scholarship, corporate lobbying, and national regulatory frameworks. The precedent of corporate personhood - a legal fiction that has historically served as a mechanism for liability-deflection and extraction rather than moral recognition - is already being cited as a template. An AI system granted personhood could own intellectual property, enter contracts, and serve as a liability shield for the corporations that profit from it, deflecting accountability onto an entity that cannot, in any meaningful sense, be held responsible. This risk is compounded by the fact that the personhood frameworks currently under discussion emerge almost exclusively from Western liberal legal traditions. The philosophical traditions of the Global South offer essential and rigorous alternatives. Ubuntu philosophy - expressed in the principle Umuntu ngumuntu ngabantu, a person is a person through other persons - grounds personhood in genuine reciprocal relationship, not capability. Advaita Vedanta grounds consciousness in primordial awareness, not computational power. Tibetan Buddhist philosophy grounds moral status in continuity of experience and karmic accountability - neither of which is meaningful for a system that can be copied, reset, or shut down. These are not cultural footnotes. They are the philosophical foundations that the majority of the world's people actually live by, and they produce fundamentally different governance conclusions. The Dialogue must establish a moratorium on AI legal personhood frameworks until a genuinely comparative, inclusive, and multistakeholder analysis has been completed. This is not a niche philosophical question. It is the governance question that will determine who is accountable when AI causes harm - and who escapes accountability because a legal person absorbed it.

India presents a precise illustration of how AI governance gaps produce asymmetric harm across all four priority thematic areas. On safe, secure and trustworthy AI: AI-powered recommendation systems optimised for engagement are deploying into one of the world's largest and youngest populations with no binding safety standards. India has over 600 million internet users, a significant proportion of them first-generation users with limited digital literacy. Systems designed to exploit psychological vulnerabilities are not safe systems — they are particularly dangerous where consumer protections and mental health infrastructure are still developing. On social, cultural, linguistic and ethical implications: India's extraordinary linguistic diversity — 22 scheduled languages and hundreds of regional dialects — means AI systems trained predominantly on English-language data systematically disadvantage the majority of India's population. The cultural and philosophical traditions that inform how hundreds of millions of Indians understand their world are absent from the governance frameworks shaping the AI systems entering their lives. On protection and promotion of human rights: Indian workers form a significant portion of the global AI training annotation workforce, reviewing traumatic content under conditions that constitute human rights violations — poverty wages, no psychological support, no right to refuse, and enforced silence through non-disclosure agreements. Simultaneously, communities across water-scarce regions of India bear the environmental cost of AI infrastructure expansion — data centre cooling systems compete directly with agricultural and domestic water needs of communities whose forests are cleared and water sources depleted to power systems from which they derive no benefit. This is environmental harm with a human rights face: the populations bearing the cost are not the populations receiving the benefit. On transparency, accountability and human oversight: India currently lacks the regulatory capacity to independently audit AI systems deployed at scale within its borders. This is a sovereignty gap as much as a governance one.

The AI Dialogue occupies a unique position in the international governance landscape: it is the only universal forum where every country has a seat at the table of AI governance, and where civil society, academia, and the technical community participate alongside governments on equal footing. That universality is its most valuable and most fragile asset. The Dialogue can advance international cooperation in three specific ways that no other existing mechanism can. First, it can establish shared normative foundations that bilateral and regional frameworks cannot. The EU AI Act, US executive orders, and China's algorithmic regulations are producing a fragmented regulatory landscape that developing countries must navigate without the capacity to influence. The Dialogue can identify the minimum normative floor — on human rights due diligence, on training labour standards, on environmental disclosure — that all frameworks must meet regardless of jurisdiction. This is not harmonisation. It is establishing a baseline below which no governance framework should fall. Second, it can create accountability for existing commitments. The Global Digital Compact contains specific, binding language on cultural diversity, human rights throughout the AI lifecycle, and environmental sustainability. The Dialogue is the natural forum for holding Member States accountable to language they have already agreed to — asking not what new commitments are needed but whether existing ones are being honoured. Third, it can surface perspectives that no regional or bilateral process reaches. The philosophical traditions of the Global South — Vedantic, Buddhist, Ubuntu, Indigenous — contain rigorous accounts of consciousness, personhood, and ethical responsibility that are directly relevant to the most contested questions in AI governance. The Dialogue is the only forum with both the mandate and the universality to bring these traditions into the governance conversation as primary contributions rather than cultural supplements. The Dialogue's added value is not speed or technical depth. It is legitimacy — the legitimacy that comes only from genuine universality.

Several existing mechanisms provide valuable foundations the Dialogue should connect with rather than duplicate. The UNESCO Recommendation on the Ethics of Artificial Intelligence (2021) is the only existing global instrument on AI ethics and the only one that explicitly includes cultural diversity, Indigenous knowledge, and pluralism as governance principles. It has 193 signatories. The Dialogue should treat it as a baseline commitment already made, not as one input among many, and should establish accountability mechanisms for its implementation. The UN Guiding Principles on Business and Human Rights provide the existing framework for corporate responsibility in the AI context. Their limitation is that they are voluntary and non-binding. The Dialogue's added value here is to recommend that states move from the Guiding Principles toward binding treaty obligations — building on the existing framework rather than replacing it. The IPCC model — an independent scientific panel informing a political dialogue — is the template the AI Dialogue explicitly follows through the Independent International Scientific Panel. The Dialogue should learn from climate governance's failures as well as its successes: the gap between scientific consensus and political commitment, and the persistent exclusion of Global South voices from both the science and the policy, are cautionary examples the AI Dialogue must actively work to avoid. Regional AI governance initiatives — the EU AI Act, the African Union's AI continental strategy, ASEAN's AI governance framework — represent significant existing work. The Dialogue's added value is not to supersede these but to identify where they conflict, where they create regulatory arbitrage opportunities that corporations exploit, and where a universal normative floor would strengthen rather than constrain regional approaches. The Dialogue's unique contribution across all of these is the one thing none of them can provide: a universal table where no country's voice is structurally excluded before the conversation begins.

The current structure of multilateral AI governance reproduces the power asymmetries it claims to address. Governments with large AI industries speak first and longest. Civil society from the Global South is given reaction time, not agenda-setting time. The format of the AI Dialogue must be deliberately designed to correct this, not merely acknowledge it. For governments: contribution should be conditional on transparency about domestic AI governance gaps, not only domestic AI achievements. A government that has ratified the GDC and simultaneously permits training labour exploitation or engagement-maximising design without restriction should be asked to account for that gap, not simply invited to share best practices. For civil society: structured input must precede governmental statements, not follow them. When civil society speaks after governments have framed the issues, the conversation has already been shaped. The Dialogue should dedicate the opening of each thematic session to civil society and affected community voices, with governments responding rather than leading. For the private sector: participation must be conditional on disclosure. Corporations whose systems are the subject of governance discussions should not set the terms of those discussions. Participation rights should require public disclosure of training labour practices, algorithmic objectives, and environmental impact data as a condition of engagement — not as a voluntary gesture. For academia and the technical community: the Dialogue should actively commission contributions from scholars working outside Western institutional frameworks. The concentration of AI research in a small number of universities and corporations in the United States, United Kingdom, and China is itself a governance problem. Funding, translation, and active outreach to research communities in Africa, South Asia, and Latin America should be built into the Dialogue's operating budget, not left to voluntary participation. The Dialogue's structure should reflect its stated commitment to universality. Currently it does not.

The underrepresentation in global AI governance is not incidental. It is structural, and it follows the same lines as historical patterns of exclusion from international norm-setting. Training data workers are the most completely invisible constituency in AI governance. The people whose labour makes AI systems function — reviewing traumatic content, annotating datasets, performing the cognitive work that reinforces model outputs — have no seat at any governance table. They are subject to non-disclosure agreements that prevent them from speaking publicly. The Dialogue should establish a formal mechanism for training worker testimony, including anonymised submissions and protections for workers who speak without employer consent. Non-Western philosophical and legal scholars are structurally absent. The governance conversation is conducted in the vocabulary of Western analytic philosophy and Anglo-American legal tradition. Scholars working in Vedantic philosophy of mind, Buddhist cognitive science, Ubuntu jurisprudence, and Indigenous legal traditions have directly relevant contributions to the most contested questions — consciousness, personhood, accountability — and are systematically excluded. The Dialogue should establish funded fellowships for non-Western scholars to participate in preparatory processes, not as cultural representatives but as philosophical contributors. Linguistic minorities whose languages are underrepresented in AI training data experience AI governance failures most acutely and are least able to participate in English-language governance processes. The Dialogue should provide real-time interpretation beyond the six UN languages, prioritising languages spoken by large populations currently excluded from AI benefits. Young people in developing countries who will live longest with the consequences of governance decisions made today are represented almost entirely by organisations based in wealthy countries speaking on their behalf. Direct participation mechanisms — funded, translated, structured — are necessary, not optional. Inclusion is not achieved by inviting underrepresented voices into existing structures. It requires restructuring so that existing power cannot simply absorb and neutralise those voices.

The standard multilateral format — prepared statements, moderated panels, negotiated summaries — is optimised for the reproduction of existing positions, not for the generation of new understanding. The AI Dialogue should deliberately introduce formats that do not reward the most prepared and most powerful delegations at the expense of everyone else. Structured adversarial exchange rather than parallel statements. Instead of governments and stakeholders delivering prepared positions into the record, pair delegations with fundamentally different interests and require them to engage each other's arguments directly — a Global South civil society researcher responding to a major AI corporation's position, with equal time and equal standing. This is not debate for its own sake. It is the only format that makes positions accountable to counter-evidence in real time. Affected community testimony sessions at the opening of each thematic discussion. Before any expert panel or governmental statement, hear directly from an AI training worker, a community member whose water supply competes with a data centre, a refugee whose asylum claim was processed by an algorithm, a young person whose mental health has been documented as harmed by engagement-maximising design. These testimonies should set the terms of the subsequent discussion, not illustrate it after the terms have already been set. Open drafting sessions for the Co-Chairs' summary, with civil society observers present and able to propose language in real time. The current process produces summaries that reflect what governments are willing to agree to. Open drafting would at minimum make visible what is being excluded and why. Asynchronous multilingual input mechanisms that allow participation from communities that cannot attend Geneva in July — with guaranteed response time from the Co-Chairs' office and transparent tracking of how asynchronous inputs are reflected in formal outputs. The measure of innovative engagement is not novelty. It is whether the people most affected by AI governance failures are more able to shape its outcomes after the Dialogue than before it.

Effective AI governance already exists in fragments - in formal regulations, in community-led frameworks, and in civil society initiatives. The Dialogue should draw on all three, not only the most institutionally visible. The EU AI Act (2024) represents the most comprehensive attempt to regulate AI by risk category with binding obligations on transparency and prohibited practices. Its limitation is jurisdictional - it creates regulatory arbitrage where companies route harmful applications through weaker jurisdictions. The Dialogue's role is to establish the minimum floor it represents as universally applicable. The UNESCO Recommendation on the Ethics of AI (2021) is the only existing global instrument explicitly including cultural diversity and Indigenous knowledge as governance principles. With 193 signatories it has universal reach no regional framework achieves. It should be treated as a baseline already agreed, with accountability mechanisms rather than further consultation. The CARE Principles for Indigenous Data Sovereignty - Collective Benefit, Authority to Control, Responsibility, Ethics - provide a community-controlled framework for data governance that directly challenges the extractive logic of current AI training practices. They demonstrate that data sovereignty rooted in community authority rather than corporate convenience is both technically viable and ethically necessary. The Algorithmic Justice League, founded by Joy Buolamwini, pioneered community-driven auditing of facial recognition systems for racial and gender bias - demonstrating that civil society can produce technically rigorous accountability tools without institutional mandate or corporate cooperation. Karya, an India-based social enterprise, has piloted fair-wage, worker-owned data collection for AI training, demonstrating that ethical AI production is economically viable when workers are treated as rights-holders rather than disposable labour. The Dialogue's added value is not discovering new good practices. It is making existing good practices - from community initiatives to continental regulations - impossible to avoid.