Ada Lovelace Institute calls for global AI governance floor at UN talks

TL;DR:

• The Ada Lovelace Institute has published its submission to the UN Global Dialogue on AI Governance, opening in Geneva this July, proposing a “global AI governance floor” — a set of minimum expectations grounded in existing international human rights law.
• The framework has three layers: substantive minimum content (what AI must and must not do), institutional infrastructure (a shared evidence base on AI risks and harms), and process commitments (regular review, state reporting, civil society participation).
• Resultsense view: this is a notably operational submission from a UK think tank — not a slogan, but a concrete structural proposal at a moment when governance is already fragmenting, with the EU diluting its AI Act this week and the US tightening pre-deployment evaluation through Commerce CASI.

The submission positions the UN as uniquely placed for the work because it brings together 193 Member States and other stakeholders, holds the international human rights mandate, and carries responsibility for international peace and security — a remit that, Ada argues, makes the UN the only forum that can speak to AI governance across both stable and fragile political contexts.

Why a floor, not a ceiling

The Institute argues that AI systems and supply chains are global while regulation is national, and that this asymmetry creates space for regulatory arbitrage and a race to the bottom. A “floor” — distinct from full harmonisation — establishes minimum thresholds while leaving jurisdictions room to adopt stricter rules above them. Its legitimacy depends on how inclusively it is built, including representation from the Global Majority in working-group leadership, not just observer status.

What the three layers mean in practice

Layer one is substantive: agreed content on rights protections and on what AI must and must not do. Ada flags how the four thematic clusters proposed by the Geneva co-chairs are defined matters — for example, whether “safe, secure, and trustworthy AI” covers existing harms and not only catastrophic risks. Layer two is institutional: a shared, routinely collected evidence base, ideally housed within an expanded UN Scientific Panel on AI mandate, to keep the floor responsive rather than speculative. Layer three is procedural: review cycles, state reporting and structured civil society participation.

UK angle and the polling backdrop

Ada cites its own UK polling — 91% of the UK public say AI should treat people fairly, 72% would feel more comfortable if AI were regulated, and 84% believe government prioritises large tech companies over citizens — and a 2025 KPMG global trust survey showing 70% support for AI regulation. The implication: there is a domestic mandate for international action, not just a technical case.

Looking forward

The Geneva Dialogue runs in July, with a follow-up scheduled for New York in 2027. The substantive work between the two — what working groups are mandated, how civil society participates, whether evidence collection is institutionalised — will determine whether the global floor becomes operational or remains a thoughtful draft. UK businesses navigating compliance across the diverging UK, EU and US regimes have a direct interest in how interoperable the result turns out to be.