AIccountable

The first Global Dialogue would succeed if it establishes concrete mechanisms for ongoing, meaningful participation by workers, trade unions, and civil society — not as afterthought consultees, but as co-architects of AI governance. Too often, governance conversations about AI happen to workers rather than with them. Specifically, AIccountable would consider the Dialogue a success if it achieves the following goals. A commitment to worker-centric governance frameworks that mandate consultation with employees and their representatives across the entire AI lifecycle, from procurement and deployment to monitoring and decommissioning. AI governance cannot remain an exclusively top-down affair between governments and technology companies while the people most directly affected by algorithmic decision-making in workplaces are absent from the table. The Dialogue should produce actionable capacity-building commitments that go beyond general digital literacy to include AI fluency programmes designed for workers, staff representatives, and trade unions, equipping them to participate substantively in governance processes, not merely to consume AI tools uncritically. This means adaptive, rights-aware learning that develops critical expertise rather than compliance-oriented training. It should affirm that AI governance is inseparable from human rights, labour rights, and democratic co-determination. The Dialogue should explicitly acknowledge that accountability mechanisms must apply not only to high-risk AI systems in the abstract, but to the concrete workplace contexts where algorithmic management, surveillance, and automated decision-making reshape power relations daily. That is the meaning of AI for emancipation. Finally, success means the Dialogue avoids becoming another declaration without institutional teeth. It should establish clear follow-up mechanisms, mandate regular reporting, and create structured pathways for non-state actors, particularly from the Global South and from organised labour, to contribute to the Scientific Panel's assessments and to shape future sessions with equal standing.

• Protection and promotion of human rights
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Transparency, accountability, and human oversight
• AI capacity-building

AIccountable's mission sits at the intersection of AI governance, ethics, and social accountability, with a particular focus on empowering workers, trade unions, and their representatives to shape how AI is deployed. These four priorities directly reflect that mission and the understanding that it is not enough to address AI risks, but rather AI has to be shaped to contribute to emancipation. Protection and promotion of human rights is foundational. AIccountable operates from a comprehensive, rights-based approach grounded in the conviction that AI governance must protect not only civil and political rights, but also economic, social, and cultural rights, including labour rights, the right to decent work, and co-determination. Without human rights as the non-negotiable floor, governance frameworks risk becoming technical compliance exercises that legitimise harmful deployments. Transparency, accountability, and human oversight is, quite literally, in our name. "AIccountable" exists because accountability cannot be an aspirational principle, it must be operationalised through genuine human oversight, meaningful "workers-in-the-loop" mechanisms, and enforceable audit requirements. This applies across the AI supply chain: from development to workplace deployment. AI capacity-building is central to our practice. AIccountable delivers AI literacy and fluency programmes specifically designed for workers, staff representatives, unions, and non-profits. We have seen first-hand that governance without capacity is governance without agency. Workers cannot exercise co-determination rights over algorithmic systems they do not understand. Capacity-building must be rights-aware, adaptive, and worker-centric, not merely technical upskilling. The social, economic, ethical, cultural, linguistic and technical implications of AI reflect the breadth of impact we engage with daily. Through our AIccountable Framework, spanning individual responsibility, ethical development, workplace co-determination, economic policy, political regulation, and societal values, we address AI's implications across all these dimensions. The world of work is where these implications converge most tangibly: jobs, conditions, dignity, and power are all reshaped by AI deployment decisions.

Yes. Several cross-cutting issues deserve explicit recognition in the Dialogue's agenda. Workplace co-determination and labour rights in AI governance remain conspicuously under-addressed in international frameworks. Workers are not merely "end users" or "affected persons", they are rights-holders with legitimate claims to participate in decisions about algorithmic systems that manage, monitor, evaluate, and discipline them. The Global Dialogue should establish workers' co-determination as a governance principle, not subsume it under general "stakeholder engagement". Collective bargaining, works councils, and trade union participation must be recognised as essential governance infrastructure for AI in the workplace. The power asymmetry between AI developers and deployers on one hand, and workers and communities on the other, is a structural issue that cuts across every thematic area. Without addressing the concentration of power in AI supply chains, including the opacity of proprietary models, the precarity of data workers, and the extractive dynamics of compute infrastructure, governance frameworks will struggle to deliver meaningful accountability. The concept of emancipatory AI, AI that complements, supports, and enriches human work and creates new opportunities rather than merely optimising labour costs, deserves recognition as a governance objective. The Dialogue should move beyond risk mitigation to articulate a positive vision of AI that serves human flourishing, democratic participation, and climate justice. Institutional investors and pension funds increasingly deploy AI in investment decisions and governance processes, yet this dimension of AI governance remains largely invisible in multilateral discussions. AIccountable's work with pension fund trustees highlights the need for governance frameworks that address AI in fiduciary and institutional contexts, not only in consumer-facing or public-sector applications. Finally, linguistic diversity in AI systems, particularly the dominance of English-language training data, risks deepening existing inequalities and marginalising non-English-speaking communities from AI's benefits.

The most immediate governance gap affecting our sector is the absence of enforceable worker participation rights in AI deployment decisions. Algorithmic management systems are being introduced into workplaces at pace, yet most governance frameworks treat this as a matter of corporate self-regulation rather than democratic co-determination or collective bargaining. AIccountable works with trade unions, staff representatives, and institutional actors such as pension fund trustees in Switzerland and across Europe. We see daily how workers and their representatives lack both the legal standing and the practical capacity to challenge or shape AI systems that directly affect hiring, scheduling, performance evaluation, and termination. A second significant gap is the disconnect between AI literacy initiatives and rights awareness. Most capacity-building programmes focus narrowly on technical skills or tool adoption. They rarely equip workers to interrogate the governance, ethical, and legal dimensions of AI systems. This creates a fluency deficit that undermines accountability: if the people most affected by algorithmic decisions cannot critically evaluate those systems, oversight mechanisms remain hollow regardless of how well they are designed on paper. On the opportunity side, the growing interoperability conversation is promising. The EU AI Act, the Council of Europe Convention, and national frameworks are creating a patchwork that urgently needs coordination. The Global Dialogue could serve as a crucial connector, particularly for actors in smaller jurisdictions like Switzerland where occupational pension governance, collective agreements, and AI regulation intersect in ways that no single framework adequately addresses. Institutional investors and pension funds represent an underexplored opportunity. AIccountable's work with employee-side pension fund trustees highlights how fiduciary governance can become a lever for responsible AI deployment, integrating ESG considerations with AI accountability. If the Dialogue recognises institutional governance as a site of AI impact, it opens pathways for accountability that bypass the traditional regulator-versus-industry dynamic entirely.

The AI Dialogue's most valuable role is one that no other existing forum can credibly fill: providing a universal platform where the Global South, organised labour, civil society, and smaller states participate on equal footing with the governments and corporations that currently dominate AI governance conversations. The G7, OECD, and AI Action Summit processes have produced useful outputs, but their membership and power dynamics inevitably shape whose priorities are centred. The Dialogue can correct this structural imbalance. Concretely, the Dialogue should function as a coordination mechanism that maps, connects, and identifies gaps between existing governance frameworks rather than duplicating them. The EU AI Act, the Council of Europe Convention, national AI strategies, and sectoral regulations like collective agreements all operate in parallel with limited mutual awareness. AIccountable's cross-border work reveals how actors operating across jurisdictions constantly navigate conflicting requirements and uneven protections. The Dialogue could commission the Scientific Panel to produce comparative assessments of governance approaches, identifying where interoperability is achievable and where genuine conflicts require political resolution. The Dialogue should also institutionalise the participation of workers and trade unions as a distinct stakeholder category, not folded into a generic "civil society" basket. The International Labour Organization has deep expertise in tripartite governance, and the Dialogue should draw on this model to ensure that those whose livelihoods are most directly reshaped by AI have structured, recurring access to the process. Finally, the Dialogue can advance cooperation by making governance knowledge genuinely accessible. This means multilingual resources, capacity-building programmes tailored to non-expert participants, and funding mechanisms that enable organisations from the Global South and from the labour movement to attend and contribute substantively rather than symbolically. International cooperation on AI governance will remain performative unless participation is materially supported, not merely formally permitted.

Several existing mechanisms are directly relevant to the Dialogue's mandate and should be actively connected rather than merely acknowledged. The International Labour Organization's work on AI and the world of work is indispensable. The ILO's tripartite model of governance, bringing together governments, employers, and workers, offers a tested framework for ensuring that AI governance includes the people most affected by algorithmic systems in workplaces. The Dialogue should formally integrate ILO expertise and adopt tripartite principles for its own stakeholder engagement structures. The Council of Europe Framework Convention on AI, Human Rights, Democracy and the Rule of Law represents the most advanced binding international instrument on AI governance. The Dialogue should treat it as a reference point for rights-based governance and explore how its principles can inform governance approaches in jurisdictions beyond Europe, particularly regarding transparency and accountability obligations. But also make it binding for private businesses. UNI Global Union and its sectoral affiliates, including UNI Europa ICTS, have developed practical policy positions on algorithmic management, workers' data rights, and AI in collective bargaining. These represent lived governance experience from the labour movement that the Dialogue should draw upon systematically. AlgorithmWatch and similar civil society organisations have pioneered practical tools for algorithmic accountability, including workplace-focused methodologies that AIccountable has directly collaborated on in Switzerland. Their applied research bridges the gap between governance principles and operational reality. The OECD AI Policy Observatory provides valuable cross-country comparison data, though its membership skews toward wealthier nations. The Dialogue's added value lies precisely in extending this analytical work to contexts the OECD does not adequately cover. The Dialogue's unique contribution is integration. None of these mechanisms currently speak to each other in a structured way. The Dialogue can make these connections legible, coherent, and actionable for practitioners who operate across institutional boundaries daily.

The Dialogue's format should break with the familiar pattern of multilateral events where governments speak, industry exhibits, and civil society observes from the margins. Meaningful multistakeholder participation requires structural changes, not just open invitations. The Dialogue should adopt a modified tripartite model inspired by the ILO, recognising workers and trade unions as a distinct stakeholder category alongside governments, the private sector, academia, and civil society. Collapsing labour into a generic "non-state actors" category obscures the specific rights, expertise, and accountability relationships that workers bring. Structured speaking time, drafting access, and follow-up roles should be allocated to each stakeholder category proportionally. The format should include practitioner-led working sessions alongside plenary debates. AIccountable's experience shows that the most productive governance conversations happen when people who actually implement AI governance frameworks in workplaces, pension funds, and union negotiations share operational insights rather than position statements. Dedicated sessions where practitioners present real cases, including failures, would generate far more useful outputs than ministerial declarations. The Dialogue should commission and publish sectoral governance assessments in advance of each session, prepared by the Scientific Panel in consultation with affected stakeholders. This gives participants a shared evidence base. These assessments should explicitly cover the world of work as a cross-cutting sector. Accessibility must be substantive, not cosmetic. This means providing financial support for participation by organisations from the Global South and from the labour movement, offering materials in all UN languages well before sessions, and creating asynchronous contribution mechanisms for organisations that cannot send delegates to Geneva or New York. The Dialogue should publish a concrete action tracker after each session, documenting commitments made and progress achieved. Without follow-up accountability, the Dialogue risks becoming a talking shop that AI's pace of development will simply outrun.

The most systematically underrepresented voices in global AI governance are those of workers and their organised representatives. This is remarkable given that the workplace is where AI's impact on human autonomy, dignity, and livelihoods is felt most directly and most frequently. Warehouse workers managed by algorithmic scheduling, call centre employees monitored by sentiment analysis, gig economy workers rated by opaque scoring systems: these people experience AI governance failures daily, yet they are largely absent from the forums designing responses. Trade unions and staff representatives bring a perspective that no other stakeholder category can substitute. They possess collective knowledge about how AI systems operate in practice, not in theory or in marketing materials. They understand the power dynamics that shape deployment decisions. And they have institutional structures for aggregating individual experiences into systemic analysis. Including them requires more than issuing invitations. It requires funded participation, structured roles in drafting processes, and recognition that labour rights expertise is as essential to AI governance as technical or legal expertise. Data workers and content moderators represent another critically absent voice. The people who label training data, review harmful content, and perform the hidden labour that makes AI systems functional are overwhelmingly located in the Global South and work under precarious conditions. Their perspective on AI governance is indispensable and almost entirely unheard. Employee-side pension fund trustees and institutional investor representatives are similarly overlooked. AI is increasingly embedded in investment decision-making and fiduciary governance, yet these actors rarely appear in AI governance discussions despite managing assets that affect millions of beneficiaries. Inclusion requires concrete measures: dedicated funding streams for underrepresented stakeholders, quota-based participation structures, asynchronous and multilingual consultation processes, and a standing commitment to hearing from the people who live with AI's consequences before hearing from those who profit from its deployment.

The most impactful format innovation would be structured "governance reality checks" where practitioners present actual AI deployment cases, including the governance decisions made, the stakeholders consulted or excluded, and the outcomes observed. These are not hypothetical scenarios or curated success stories but honest accounts from workplaces, pension funds, union negotiations, and public institutions where AI governance has been tested against reality. A second format worth adopting is the "fishbowl" model, where a small group of directly affected people, such as workers subject to algorithmic management, data labellers, or content moderators, hold a facilitated conversation in the centre of the room while delegates listen before joining. This inverts the usual hierarchy where policymakers speak first and affected communities respond. It also makes the human stakes of governance discussions impossible to abstract away. The Dialogue should experiment with pre-session collaborative drafting, using shared digital platforms where stakeholders can comment on, amend, and challenge proposed texts. This extends meaningful participation beyond those who can afford to attend in person and produces richer outputs because points of disagreement surface early rather than erupting unproductively during time-limited plenary sessions. Regional preparatory consultations held in partnership with organisations like UNI Global Union, the ILO, and national trade union federations could feed worker perspectives into the Dialogue systematically rather than relying on whichever labour organisations happen to secure accreditation. The Dialogue should publish session recordings, working documents, and draft outputs in all UN languages within days rather than months. Transparency delayed is transparency denied, and organisations operating in non-English contexts cannot meaningfully engage with materials that arrive only in English or only after decisions have already solidified.

Collective bargaining and collective agreements represent one of the most underappreciated yet effective mechanisms for AI governance. In sectors where unions have negotiated algorithmic management clauses, workers gain enforceable rights to information, consultation, and co-determination before AI systems are deployed. Spain's "riders' law" requiring platforms to disclose algorithmic parameters to worker representatives demonstrated that labour regulation can deliver concrete transparency obligations where voluntary frameworks have not. There is a growing database with best practices from collective bargaining, e.g. in the USA, in Spain, and in other countries. The EU AI Act represents the most comprehensive regulatory approach to date, establishing risk-based obligations. Its requirements for transparency, human oversight, and conformity assessment create a governance architecture that other jurisdictions can learn from, even where direct adoption is not feasible. This approach has to be enhanced and extended to AI and algorithmic management tools at the workplace. The Council of Europe Framework Convention on AI anchors AI governance in human rights, democracy, and the rule of law, providing a binding international reference point that complements softer multilateral commitments. AlgorithmWatch CH's collaborative work with Swiss trade syndicom on empowering employees when algorithmic systems are used in the workplace offers a practical methodology for translating governance principles into actionable tools that workers and staff representatives can actually use during consultation processes. At the institutional governance level, AIccountable's work with employee-side pension fund trustees in Switzerland through PK-Netz demonstrates how fiduciary governance structures can integrate AI accountability into investment oversight and organisational decision-making. This approach treats existing democratic governance institutions as sites for AI accountability rather than waiting for entirely new regulatory bodies to be created. The common thread across these examples is that effective AI governance works best when it builds on existing rights, institutions, and power structures rather than inventing parallel processes that lack enforcement mechanisms or democratic legitimacy.