Alignerr LLC

A successful first Global Dialogue should anchor international AI governance in three fundamental values: safety, fairness, and transparency, with safety taking precedence over speed of deployment. Translating these values into concrete outcomes would mean the following: On safety, the Dialogue should produce a shared commitment that the pace of AI development must not outrun our collective capacity to prevent harm. This should prioritise high-risk applications, with scope to extend over time as the evidence base develops. This includes recognising that harm is not evenly distributed. Vulnerable communities, and the intersections between them, are disproportionately exposed to direct and indirect risks of AI systems. A meaningful outcome would be agreement on baseline safeguards, regardless of where they are developed or deployed. This should prioritize high-risk applications but also extend to moderate risk applications in the future. On fairness, the Dialogue should affirm that when AI systems materially affect people's lives, fairness cannot be an afterthought layered onto a finished product. It must be embedded into the system design. A successful outcome would include guidance on participatory approaches that bring affected populations into the design and oversight of systems that shape their access to services, opportunities, and rights. Trade-offs in AI development are unavoidable, and that makes transparency a design choice rather than a disclosure exercise. People subject to algorithmic decisions need to understand how those decisions are reached and how to contest them through accessible override mechanisms. Crucially, transparency should also bind the institutions governing AI. Policymakers, regulators, and international bodies owe the public visibility into the choices being made on their behalf, not only the rules imposed on developers and deployers. The Dialogue will succeed if it moves these principles from aspiration toward shared, actionable commitments, and if it establishes a governance process that is itself accountable to the people AI will most affect.

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

Safe, secure and trustworthy AI is the foundation, and safety is only achievable with meaningful human oversight. Automation of consequential decisions without humans effectively positioned to intervene, audit, and override is not a safety strategy. Urgent action here means agreeing on shared expectations for AI systems and resisting pressure to let competitive pace dictate the safety threshold. In practice, this means defining minimum conditions for effective oversight: humans must have the authority, technical visibility, and institutional incentives to intervene,rather than being formally designated as oversight without the real capacity to change outcomes. There are currently no authoritative guidelines on where and how human-in-the-loop controls should be positioned within decision pipelines, which means oversight is often implemented inconsistently or too late to meaningfully affect outcomes. Transparency, accountability, and human oversight is how that foundation becomes legitimate. Affected populations need to understand how AI-driven decisions are reached and have accessible mechanisms to contest them. Accountability must extend beyond developers and deployers to include the institutions governing AI. Policymakers and international bodies owe the public visibility into the choices being made on their behalf. Interoperability of governance approaches matters because AI systems cross borders that governance does not. The current landscape is fragmented and often contradictory, and the burden of mapping conflicting obligations falls disproportionately on SMEs operating with tight budgets and limited capacity. Rules must be clear, aligned, and operationalisable to be enforceable. The Dialogue can advance shared reference points that allow distinct regulatory cultures to remain coherent with one another, so that compliance becomes a substantive commitment rather than a routing exercise. Social, economic, ethical, cultural, linguistic and technical implications anchor the entire conversation in the populations AI will actually affect. Policy that treats these as downstream considerations rather than design inputs will systematically underweight the people most exposed to harm. Multilingual access, cultural context, and economic disparities are not edge cases. They define whether AI governance serves the global majority or a narrow subset of it.

I see three cross-cutting issues not captured by the listed themes above: - Environmental and resource implications of AI: The compute, energy, and water demands of frontier AI development are growing rapidly, and the burden of that growth is not evenly distributed across regions or communities. Governance frameworks that treat sustainability as adjacent to AI policy, rather than embedded within it, will struggle to align with parallel international commitments on climate and resource equity. The Dialogue is well placed to surface this as a governance question, not only an infrastructure one. - Labour and economic displacement: this is partially captured under social and economic implications, but it deserves dedicated attention. AI is reshaping the structure of work across sectors and regions on a timeline that policy is not matching. Without proactive engagement, the costs of transition will fall on workers and economies least equipped to absorb them, while the gains concentrate elsewhere. Governance should address not only how AI is built and deployed, but how the benefits and disruptions are distributed. - Concentration of power and market structure: this underpins all the other themes and constrains what governance can realistically achieve. A small number of actors currently shape the foundational layers of the AI stack: compute, frontier models, and the data pipelines feeding them. This concentration affects safety (who decides what is safe), interoperability (whose standards prevail), and fairness (whose interests are encoded). Without addressing it, governance risks regulating the surface while the foundations remain outside democratic reach. Surfacing these issues explicitly would strengthen the Dialogue's capacity to govern AI as it actually exists, rather than as it is often described. Across all three areas, risks and costs are systematically externalised to less powerful regions and populations, while benefits concentrate elsewhere. The Dialog should treat these as governance design failures, rather than sector-specific side effects. This would allow the Dialogue to address root causes instead of symptoms.

From an EU vantage point, and particularly from the perspective of SMEs (which form the backbone of the European economy) navigating compliance, the challenges and opportunities are tightly linked. The EU AI Act has set a substantive regulatory benchmark, and its risk-based structure has an extraterritorial effect. This is an opportunity: a coherent regional framework gives organisations a structured way to think about obligations, and it has accelerated investment in governance capacity across the ecosystem. The challenge is that this framework does not exist in isolation. Organisations operating across borders must reconcile the AI Act with sectoral regulation, data protection regimes, emerging national AI laws, and international standards that are still consolidating. In practice, this fragmentation forces organisations, particularly SMEs, to act as de facto integrators of global governance, translating overlapping and sometimes conflicting requirements into internal controls without clear guidance or economies of scale. For large firms with dedicated legal and compliance functions, this is manageable. For SMEs it is a serious operational burden. Mapping overlapping obligations, interpreting ambiguous provisions, and embedding controls into workflows on tight budgets and limited capacity often pushes governance toward checkbox compliance rather than substantive risk management. This has direct consequences for the priorities raised earlier. Safety and human oversight are harder to operationalise when teams are stretched thin interpreting conflicting guidance. Transparency and accountability suffer when documentation becomes a defensive exercise rather than a tool for affected populations. And the social, economic, and cultural implications of AI are unevenly addressed when only well-resourced organisations can afford to engage with them seriously. The opportunity for the Dialogue is to support interoperability that is genuinely operationalisable: shared reference points, mutual recognition where appropriate, and practical guidance that lowers the cost of doing governance well. A framework that only the largest actors can afford to comply with is not, in practice, a framework for safe and trustworthy AI.

The Dialogue's distinctive value lies in what it can do that no single jurisdiction or industry forum can: convene a genuinely global conversation under a mandate that includes the states and communities most often absent from AI governance discussions. Three roles seem particularly important: First, the Dialogue can serve as a bridge between fragmented regulatory regimes. It will not, and should not, attempt to harmonise every framework into a single instrument. But it can advance shared reference points, common vocabulary, and points of mutual recognition that allow distinct regulatory cultures to remain coherent with one another. This is the practical foundation of interoperability, and it directly addresses the compliance burden currently borne by smaller actors operating across borders. To be effective, this bridging function should produce concrete outputs, such as mapped equivalences between key obligations and clearly defined conditions for mutual recognition, rather than remaining at the level of high-level alignment. Second, the Dialogue can broaden the base of who is represented in international AI governance. Much of the existing conversation is shaped by a small number of jurisdictions and a smaller number of firms. A UN-anchored process is well placed to bring in states, civil society organisations, and affected communities whose perspectives are essential for governance to be legitimate, particularly on the social, economic, cultural, and linguistic dimensions raised in the listed themes. Inclusion here changes what governance prioritises. Third, the Dialogue can serve a transparency function for international AI governance itself. By keeping stakeholders informed about the decisions being taken on behalf of affected and vulnerable populations, and by creating channels for those communities to shape those decisions rather than only receive their consequences, the Dialogue can help close the legitimacy gap that currently sits between AI governance institutions and the people they govern. If the Dialogue does these three things well, it will not replace existing efforts. It will make them more coherent, more inclusive, and more accountable to the people AI affects most.

I propose two possibilities to make existing initiatives more coherent and more accountable: The first is a standing interoperability mechanism linking the Dialogue with regional and regime-specific frameworks, including the OECD AI Principles, the Council of Europe Framework Convention on AI, the EU AI Act, the ASEAN Guide on AI Governance, the African Union Continental AI Strategy, and equivalent instruments. Its mandate would be to produce a living map of obligations across these regimes, identify points of convergence and contradiction, and recommend mutual recognition where feasible. This would draw on the technical work of ISO/IEC and NIST to ground recommendations in auditable practice. To ensure this mechanism is operationally useful, outputs could be maintained as a publicly accessible, machine-readable obligations registry with version control and clear ownership, allowing organisations (particularly SMEs) to directly integrate evolving requirements into their compliance workflows rather than interpret them ad hoc. The added value: the current fragmentation imposes a disproportionate compliance burden on smaller actors and lower-capacity jurisdictions, and a structured interoperability function would translate the Dialogue's mandate into something operationally meaningful. The second is a formal civil society and affected communities channel feeding into the Dialogue. Building on the normative foundation of UNESCO's Recommendation on the Ethics of AI and the multistakeholder tradition of the ITU's AI for Good, the Dialogue would establish a standing input mechanism for civil society organisations, worker representatives, and communities most affected by AI deployment, with particular attention to the Global South. This could be a structured process with defined timelines, transparent submission procedures, and an obligation on the Dialogue to publicly respond to inputs received. The added value: governance shaped only by states and well-resourced industry actors cannot credibly claim to serve the populations AI most affects.

It's crucial that the Dialogue allows the participation of stakeholders whose voices are typically harder to surface in international forums. Civil society organisations and affected communities should have a structured input role rather than observer status. This means defined submission processes, transparent timelines, and obligations on the Dialogue to publicly respond to inputs. Particular attention should go to organisations representing populations most exposed to AI harms, including workers, marginalised communities, and those in lower-capacity jurisdictions. The private sector, including SMEs and not only large firms, should contribute on operational realities: where compliance is feasible, where it is not, and where regulatory ambiguity is creating risk-averse outcomes that limit beneficial use. SME perspectives are particularly important because they reveal what governance looks like at the capacity level most organisations actually operate at. The technical community and researchers should provide independent expertise, distinct from industry positions, on capabilities, limitations, and emerging risks. States should engage not only through formal interventions but through structured peer exchange on implementation experience, including what is working, what is not, and where regulatory choices are creating unintended burdens. International organisations and standards bodies should contribute technical grounding and translation capacity, helping convert principles into auditable practice and ensuring that the Dialogue's outputs connect with existing instruments rather than sitting alongside them. In terms of structure, the Dialogue should favour interactive working sessions over plenary statements, publish its evidence base and decision logic openly, and maintain continuity between annual sessions through standing working groups. Episodic consultation will not produce the depth of engagement the mandate requires. Sustained, structured, and transparent participation will.

Indigenous communities and linguistic minorities are largely absent from conversations about systems that increasingly mediate access to information, services, and cultural representation. Their inclusion matters not only as a matter of equity but because their perspectives surface harms that dominant frameworks systematically miss. Intersectional populations, particularly women of color from the Global South, remain underrepresented in technical and policy leadership shaping AI governance, despite being disproportionately affected by many of its harms. Global South states and stakeholders are frequently positioned as recipients of governance frameworks rather than co-authors of them. Meaningful inclusion requires travel and participation support, agenda-setting power, and recognition that capacity gaps are not only technical but also relate to who gets to define what counts as a problem worth governing. SMEs and civil society organisations operating with limited resources are underrepresented relative to large firms and well-funded advocacy organisations. The participation cost of international processes filters out exactly the actors whose operational realities the Dialogue most needs to understand. Communities directly affected by AI deployment, including workers facing automation, recipients of algorithmic decisions in welfare and immigration systems, and populations subject to AI-enabled surveillance, are rarely present in governance forums where decisions about them are taken. Inclusion requires funded participation, accessible formats, and structured channels through which their input materially shapes outcomes rather than being noted and set aside. Inclusion across these groups requires funded participation that removes the cost barrier, structured input mechanisms with transparent response obligations, and agenda-setting roles instead of temporal consultation roles.

Structured working sessions on specific governance challenges, organised around concrete problems rather than broad themes to achieve substantive problem-solving. Sessions on topics such as cross-border incident response, mutual recognition of conformity assessments, or operationalising human oversight to surface where convergence and disagreements lie. Scenario-based exercises, drawing on the Scientific Panel's evidence base, could help stakeholders work through how different governance approaches would handle specific cases. This grounds abstract debate in tangible consequences and reveals assumptions that position statements often hide. Regional and cross-regional caucuses would allow states and stakeholders with shared contexts to develop coordinated positions, while cross-regional sessions would build understanding across divides. This is particularly important for ensuring Global South perspectives enter the Dialogue as equally strong positions to those of more developed economies. Civil society and affected community panels integrated into the main programme, not relegated to side events, would ensure that the populations AI most affects are present when commitments are shaped. Their interventions should be scheduled with the same weight as state interventions. Open documentation of the Dialogue's evidence base, working drafts, and decision logic, accessible in multiple languages, would allow stakeholders unable to attend in person to engage substantively with the process. Asynchronous participation channels with defined response obligations would extend the Dialogue beyond the room. Continuity mechanisms between annual sessions, including standing working groups with defined deliverables, would prevent the Dialogue from resetting each year.

The EU AI Act demonstrates a risk-based structure that calibrates obligations to potential harm, with specific provisions for high-risk systems, transparency requirements for general-purpose AI, and prohibitions on practices considered incompatible with fundamental rights. Its implementation experience, including the challenges of operationalising it for SMEs, offers practical lessons regardless of whether other jurisdictions adopt the same model. The NIST AI Risk Management Framework provides a voluntary, sector-agnostic approach to identifying and managing AI risks, designed for adaptability across organisational contexts. Its accessibility makes it particularly useful as a reference for actors without dedicated compliance functions. UNESCO's Readiness Assessment Methodology helps states evaluate their institutional, regulatory, and technical readiness for ethical AI governance, providing a structured entry point for jurisdictions building governance capacity from a lower starting base. The Council of Europe Framework Convention on AI represents the first binding international treaty on AI, anchored in human rights, democracy, and the rule of law. It demonstrates that meaningful international agreement on AI governance is achievable. Algorithmic impact assessment requirements, including those in Canada's Directive on Automated Decision-Making and emerging practice in several jurisdictions, offer a concrete mechanism for translating principles like transparency and accountability into specific obligations on public sector AI use. Participatory governance practices, including community oversight boards for facial recognition deployments in several US cities and worker consultation requirements in some European jurisdictions, demonstrate that affected populations can be structurally embedded in governance rather than consulted retrospectively.