Ada Lovelace Institute

Long-term, the Global Dialogue has the opportunity put in place a set of minimum expectations that apply to all AI systems globally: expectations designed to ensure that the development and deployment of AI uphold the dignity, rights and security of all people; advance inclusive and equitable development; preserve and promotes the conditions for peace; and advance the interoperability different AI governance regimes. We call this the 'AI governance floor'. The success of this first Global Dialogue hinges on its ability to establish the structures required to shape this AI governance floor collectively across Member States and other stakeholders, including the public, leading up to and including the 2027 Dialogue. The 2026 Dialogue will be a success if it achieves five concrete outcomes: a) Agreement on the substantive issues (thematic clusters) and their scope that must be addressed (including their relationship) as part of the governance floor b) Clarity on process ownership for the period leading up to and including the 2027 Dialogue (working groups, co-chairs, etc.) c) Cooperation and complementarity with the UN Scientific Panel on AI to build a shared, authoritative evidence base on AI's opportunities, risks and harms (the evidence base essential to ground governance proposals) d) A public commitment to and support for the creation of globally inclusive channels that bring public experience and expectations into the Dialogue, catalysing public deliberation on Dialogue themes and outcomes e) Facilitation of the 'Dialogue of dialogues', driving the complementarity of other efforts and processes that share the ambition to shape global AI governance. The outcome document of the Global Dialogue will be successful if it captures these five concrete outcomes, establishes clear processes and the mandates to drive progress on substantive issues, and serves as the stepping stone to ensure continuity and accountability for the Dialogue series.

• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Interoperability of governance approaches
• Protection and promotion of human rights
• Transparency, accountability, and human oversight

We recommend prioritising these four thematic areas because the UN Global Dialogue on AI should focus its energy on thematic areas that (a) can only be (fully) resolved at global level and (b) can only be led by the UN (incl. its family of institutions) with its Charter, role as the custodian for human rights and institutions, inclusive membership and legitimacy to establish shared expectations across its Member States. The four chosen thematic areas are also the substantive foundation for the AI governance floor by design: the AI governance floor consists of the elements necessary to move towards a shared future in which the global community thrives alongside AI. Understanding AI's implications across social, economic, ethical, cultural, linguistic and technical dimensions provides the evidence base without which the governance floor risks being speculative. Interoperability between governance approaches addresses the fragmentation that is already emerging between governance regimes. Transparency, accountability and human oversight are the procedural minimum to assess and evolve the governance floor in light of new evidence, and to enable rights protections. Human rights protection and promotion is not one priority among equals; it is the framework within which the other thematic areas operate. As convenor, the UN brings its Charter and Mandate including its role as the custodian of the international human rights system. The human rights grounding of this process is not a choice but the foundation; a grounding that only this Dialogue can provide for the global community translating human rights into shared minimum expectations for our future with AI. The remaining areas are important, but they can be actively advanced through other processes and actors. By hosting the 'Dialogue of dialogues', the UN can facilitate such focus and complementarity.

First, there is the risk of a narrow safety framing. With 'Safe, secure and trustworthy AI' as a theme, if the Dialogue's safety discussions are shaped by long-term or catastrophic risks, nearer-term and documented harms to individuals, communities and institutions risk being marginalised. Human rights law provides the universal framework for defining what AI safety means; the latter must be conceptualised broadly. The Dialogue should ensure that the human rights framework is the connective tissue across thematic areas not replaceable by a (narrow) safety conversation. Second, there is the risk of governance without the infrastructure for evidence and legitimacy. A governance floor that cannot be grounded in routinely, systematically collected evidence on AI's risks, harms and rights implications will remain speculative and vulnerable to capture. The UN Scientific Panel on AI's mandate must include (or be supplemented by) a structural, institutionally supported globally inclusive monitoring of AI risks and harms so that governance solutions can be targeted, adaptive and based on shared evidence. However, evidence alone is not enough. AI governance decisions shape the lives of people who are not represented in expert panels or state negotiations. Without structured mechanisms that give public voices procedural weight in the governance process, we risk a legitimacy deficit that undermines durability. Shared evidence and inclusive deliberation are the infrastructure on which governance must be built. Third, there is the risk of the use of AI in conflict and fragility affecting conditions for peace and humanitarian action. Without addressing this dimension, AI governance risks becoming a framework that protects rights in stable contexts, leaving the most vulnerable populations unprotected. The UN Charter mandate for maintenance of international peace and security and its operational role in peacekeeping in humanitarian action give this Dialogue unique standing to establish minimum governance expectations for AI in conflict-affected contexts.

Our 2025 polling and survey work shows that 91% of the UK public think it is important that AI systems are developed and used in ways that treat people fairly, 72% say that legislation would make them feel more comfortable with AI. But 84% fear that government prioritises tech over public interests. Likely, this is not a UK anomaly, but reflective of a disconnect between democratic expectations and governance reality. In the UK, AI is governed by horizontal and sectoral regulation. A gap is the absence of obligations on AI developers outside of safety contexts. Risks that arise during development are embedded and propagate down value chains across the economy, particularly for general-purpose AI models. UK post-Brexit positioning as the 'lighter-touch regime' is framed as competitive advantage, indicative of race-to-the-bottom dynamics. A governance floor makes sure the race stops before harms occur. The EU put forward its AI Act (2024), a horizonal, risk-based regulation of AI systems building on EU digital regulation including the General Data Protection Regulation (GDPR, 2016). But concerns about competitiveness and security have reopened the AI Act, the GDPR and related EU digital regulation, symptomatic of concerns about the use of compliance fragmentation for arbitrage. A governance floor reduces opportunity for arbitrage. Regardless of jurisdiction, rights protections and transparency obligations that stop at borders are not protection for all people. A global governance floor does not stop at borders. Separately, a risk with, e.g., the EU is that the framing of interoperability as adoption of its approach and standards (the 'Brussels effect') will be resisted by Global Majority countries and others. A global governance floor whose development is led by the UN is inclusive from the outset. National and regional governance reveals what global governance must solve. The UN is positioned to deliver it.

The Dialogue distinguishes itself from other efforts with the objective to shape global AI governance by its UN mandate. It can establish shared expectations that no other forum can. But if the Dialogue remains only a dialogue, it will plateau into soft coordination. Instead, it must commit to a shared objective and concrete outcomes lasting beyond the 2026 and 2027 Dialogues. That's why we propose the AI governance floor: shared expectations for the development and deployment of AI that uphold human rights, advance equitable development and preserve the conditions for peace. Expectations produced through the UN process have the highest chance of acceptance (cf. other for a/processes) because of how they were made: inclusively with all Member States and stakeholders at the table. This earns legitimacy that no other forum can claim. International cooperation requires infrastructure, not just agreement. The Dialogue can mandate the institutional components that the governance floor requires: structured, routine, inclusive evidence generation (AI risk and harm monitoring) and civil society participation channels including public deliberation. The UN must further define how it contributes, as the custodian of the AI governance floor, to driving interoperability across evolving AI governance regimes. These components, once institutionalised, become the legacy of the 2026 and 2027 Dialogues. The Dialogue provides the occasion to clarify the mandates of UN entities including but not limited to UN ODET, UNESCO, ITU, OHCHR, ILO, WIPO, UNDP regarding their respective contributions to AI global governance. This is to make the UN coherent and effective rather than fragmented in its AI contributions and ensuring that Dialogue outcomes translate into sustained institutional capacity and action beyond 2027. Finally, the Dialogue can drive complementarity and division of labour across other efforts to shape AI governance; the existence of multiple processes becomes a strength, rather than a source of fragmentation.

The risk is not too few connections but too many: a Dialogue that tries to connect with everything connects with nothing meaningfully. The Dialogue should prioritise initiatives that have either produced substantive governance content that a governance floor can build on or represent constituencies whose buy-in is essential for the floor to function effectively. Specifically: • UN Scientific Panel on AI: acknowledging the existing relationship, the Dialogue must define the mandate and division of labour on inclusive evidence generation, public deliberation and their institutionalisation beyond the 2027 Dialogue • AI Summit Series: has produced political momentum; the Dialogue can complement and put in place required infrastructure for monitoring and delivery, with future Summits potentially creating 'moments' for review of progress and political commitment • OECD (Hiroshima Process on AI): has advanced governance floor thinking (through e.g. its OECD AI principles,) amongst developed economies • Council of Europe Framework Convention on AI: the most advanced, binding instrument currently in existence for AI governance grounded in a rights-based logic • Regional bodies - EU, AU, ASEAN: representing three major governance interests and trajectories, the Dialogue can drive interoperability • CSTD Working Group on Data Governance: data governance is a precondition for AI governance; these two UN-led processes must be closely connected to prevent critical gaps and avoid redundancies • UN system contributions: notably, UNESCO (Recommendation on the Ethics of AI, the broadest soft law instrument on AI), OHCHR (human rights), ILO (labour dimensions of AI), WIPO (intellectual property), and UNDP (capacity building); the Dialogue presents an opportunity to clarify mandates. The Dialogue's proposed continuity and review structures present an opportunity for the global community to assess commitments and their progress against them collectively.

The format and structure of the Dialogue determine whose voices shape outcomes. The Dialogue must be designed for substantive progress, not symbolic inclusion. This requires meaningful involvement of people experiencing and affected by AI technologies to shape thematic progress and any processes and structures that will put in place through the 2026 and 2027 Dialogues. Specifically, the 2026 Dialogue has the opportunity to put in place working groups organised by thematic clusters. These groups need a strong mandate, timelines and deliverables to drive substantive progress in between the 2026 and 2027 Dialogue. Chairs and co-chairs of the groups must reflect geographic, gender and stakeholder diversity (i.e., the type of organisation and the constituencies they represent) including Global Majority representation in leadership. The groups should include civil society and research institutes/academia alongside Member States as substantive contributors. Regional gatherings and online formats between Dialogues maintain momentum and broaden participation beyond those who can travel to Geneva and New York; consider organising regional gatherings alongside major events (e.g. regional IGF). Partner with established public deliberation initiatives (e.g. PAIRS, Connected by Data, ISWE) to bring public voices into the Dialogue process as a mechanism to generate public input that feeds directly into working group deliberations. Recognising the challenges of traditionally underrepresented groups in these processes, a dedicated fellowship programme to increase their participation can enable participation in the lead-up and in between Dialogues. Fellows must be given the opportunity to deeply integrate themselves into the work and working groups for their meaningful participation and contribution.

Meaningful participation of different publics in the dialogue requires representation of often excluded perspectives and ways for those perspectives to shape outcomes. Here, precision matters. To be specific about constituencies: • People directly harmed by AI systems: welfare recipients, migrants, asylum seekers, and people in contact with criminal justice systems are all likely subject to automated decisions but often structurally excluded from governance conversations • Workers in AI affected labour markets: gig workers and platform workers are all likely subject to algorithmic management and face automation without adequate protection (incl. knowledge extraction for AI system development); data labellers subject to harms and extractive practices • Indigenous communities: their data and cultural heritage is appropriated by AI often without consent • Women and girls: disproportionately exposed to non-consensual imagery and algorithmic discrimination (e.g. hiring) • Children: disproportionately exposed to AI-driven content recommendation (and increasingly AI-generated content without quality controls), data collection and profiling including in educational contexts and most vulnerable to long-term epistemic and developmental harms • People in conflict and fragile contexts: at risk of being the most exposed to ungoverned AI and least represented in governance conversations • Small and developing states: invited in principle but capacity constrained in practice (limited resources to send delegations and meaningfully shape dialogue) • Public voices: publics everywhere have limited access to governance conversations. To limit their risk of exclusion, there must be: • Efforts to make AI risks and harms visible in evidence through targeted data collection efforts and disaggregated data (e.g. by gender, age, minoritised ethnicities) • Inclusion of constituency representation (labour rights organisations, women's rights organisations, youth representatives) or proxies where necessary (e.g. ICRC for people in conflict and fragile contexts, UNICEF for children) • Meaningful support for inclusive participation, e.g., funding and fellowship programmes for individuals representing groups at risk of underrepresentation, and their structural integration into thematic working groups.

We propose the plenaries are used for focused discussion structured around specific questions requiring political commitment. Where necessary, plenaries can accommodate structured contributions including prepared speeches. We recommend opening the opportunity to publish prepared statements online prior to and during the Dialogue to increase who can contribute and broaden inclusivity. Working groups organised around thematic clusters are an effective format for driving thematic progress. Their work should commence prior to the 2026 Dialogue. They can provide continuity between the 2026 and 2027 Dialogues. During the 2026 Dialogue, a live conversation in the room must be encouraged by format. Discourage prepared speeches in working groups; these can be replaced by short reaction rounds (timed reactions to proposals or evidence presented) to drive engagement with the conversation in the room. Encourage the participation of technical experts from Member States and all stakeholder groups. Throughout, we encourage the use of online tools for parallel contributions including shared screens in the room, periodic structured polling and written contribution channels to enable parallel participation. This may be particularly useful for non-native English speakers; it levels the playing field. Facilitation quality determines whether this format succeeds or fails; skilled facilitators comfortable working across parallel channels of input will be required. Connecting delegates to people experiencing the benefits and harms of AI technologies in contexts around the world is vital. This can be achieved through local in-person and online links to participatory dialogues around the world, so subject specific experts and lived experience experts can have the opportunity for exchange of perspectives. These initiatives have been demonstrated to ground governance debates in the concerns of citizens and publics.

Existing policies and practices demonstrate both what effective AI governance can achieve and where global gaps remain. The EU AI Act establishes a horizontal, risk-based framework for AI systems that has already influenced governance conversations globally. The accompanying General Purpose AI (GPAI) Code of Practice begins to address frontier model governance (although the processes for agreeing it were subject to disproportionate industry influence). The Council of Europe Framework Convention on AI provides a rights-based, binding complement. Together, these instruments demonstrate that meaningful AI governance is achievable, but their current reach and legitimacy limitations (at global level) confirm that a global floor is necessary, not a nice-to-have. Effective governance requires public legitimacy. Our work in the UK demonstrates that the public already expects a governance floor and the gap between expectation and delivery is eroding trust. The Dialogue is well positioned to start to address this gap at a global level, building on public deliberation best practices and partnering with established organisations in this space.