A decade of AI summits has produced a great deal of paper and very little coherence. Bletchley, Seoul, Paris, New Delhi; the OECD, the G7, the EU, the Council of Europe, the African Union, ASEAN. Each one did real work. Stacked together, they add up to a rulebook that no two jurisdictions read the same way. This week the UN convenes its first Global Dialogue on AI Governance in Geneva to try to stitch the mess back together, and for any UK business selling AI-enabled products beyond its own borders, the mess is not an abstraction. It is already a line on the compliance budget.
The dialogue of dialogues
The Ada Lovelace Institute, writing ahead of the inaugural UN Global Dialogue on 6-7 July, frames the problem cleanly: AI governance is “fragmented, partial and unequally effective.” That is not a complaint about too little regulation. It is a complaint about too many disconnected pieces of it.
Look at what the last few years have actually produced. Voluntary guidance from clubs of wealthy states. Binding law inside single jurisdictions. Regional declarations with no enforcement teeth. National statutes taking shape at wildly different speeds. Every instrument does necessary work in its own lane, and none of them can set a floor the others must respect.
| Instrument or process | Type | Practical reach |
|---|---|---|
| OECD Recommendation on AI (2019, 2024) | Voluntary guidance | Members and adherents |
| G7 / Hiroshima Process Code of Conduct (2023) | Voluntary commitments | G7 plus signatory firms |
| AI Summit Series (Bletchley 2023 → New Delhi 2026) | Declarations + voluntary pledges | Bletchley signed by 28 states |
| EU AI Act (2024) | Binding law | EU jurisdiction |
| Council of Europe Framework Convention on AI (2024) | Binding treaty | Ratifying states only |
| Regional frameworks (AU, ASEAN, Santiago, Montevideo) | Strategies and declarations | Regional |
| National legislation | Binding | Single state |
| UN Global Dialogue on AI Governance (2026) | Coordinating forum | Aims at all 193 members |
Critical Context: The Bletchley Declaration, hosted by the UK and treated at the time as a landmark, was signed by just 28 of the UN’s 193 member states. A framework that reaches one in seven governments is not a global settlement. It is a coalition.
The structural gap the UN is trying to fill is the one nobody else can. It has near-universal membership and the standing to set minimum expectations that apply everywhere, not just among the states that turned up to a particular summit. The track record is not nothing: the UNESCO Recommendation on the Ethics of AI (2021) was adopted across its 194 members, and two UN General Assembly resolutions on AI in 2024, led separately by China and the United States, passed by consensus of all 193 states. Getting Washington and Beijing to agree on the same page, twice, is the kind of convening only the UN does.
Why a fragmented rulebook is a business cost, not just a policy one
It is tempting for a UK operator to read all this as diplomacy that happens somewhere else. The Ada Lovelace analysis, drawing on the UN Scientific Panel on AI’s preliminary 2026 report, makes the commercial case harder to wave away.
The first cost is regulatory arbitrage. An AI system rejected in one jurisdiction on human-rights grounds can be, and already is, deployed in another where protections are thinner. That sounds like someone else’s ethics problem until you are the firm competing against a rival who shipped the thing you were told you could not.
The second cost is the compliance patchwork itself. Different national and regional rules rarely reconcile. Reconciling them is expensive, and the expense does not scale down. A multinational absorbs it; a UK small or medium-sized enterprise trying to sell into three regulatory regimes at once pays disproportionately for the privilege.
Hidden Cost: Fragmentation is quietly a competition policy. As compliance costs concentrate among the companies large enough to swallow them, the patchwork entrenches incumbents, and does so most sharply for frontier models. The rulebook nobody coordinated ends up protecting the biggest players in it.
The third cost is exposure most firms never price. Critical infrastructure increasingly runs on AI, and without interoperable arrangements for testing and disclosing cross-border risks, a system deemed unsafe anywhere can put connected systems everywhere at risk. The evidence itself is fragmented too: the UN panel found not just inconsistent rules but incomplete, geographically skewed information about where AI is failing. You cannot manage a risk your own market cannot see clearly.
The UK’s genuinely awkward position
Britain sits in an unusually exposed spot in this picture, and it is worth being honest about why.
The UK hosted the summit that produced Bletchley, which gives it a claim to leadership on frontier-AI safety. It has also chosen to stay outside the EU AI Act and pursue a lighter-touch, sector-led approach at home. Both are defensible. Together they mean UK firms operate at the seam between regimes rather than comfortably inside one. A British company selling into the EU meets the AI Act’s obligations regardless of Westminster’s preferences; selling into other markets, it meets whatever those markets decided; at home, it works to guidance that is still taking shape.
| Business function | What fragmentation changes | Exposure |
|---|---|---|
| Legal and compliance | Multiple, non-reconciled regimes per product line | High — cost scales with market count |
| Product and engineering | ”Safe here” does not travel; features gate by jurisdiction | High — rework late in the cycle |
| Sales and go-to-market | Market access blocked or delayed by local rules | Medium — slows expansion |
| Strategy | Standards set now by others become fixed constraints later | Medium — hard to renegotiate |
Strategic Reality: The UK’s out-of-the-EU stance is not the same as regulatory freedom. For any firm that exports, the binding constraint is the strictest regime it sells into, not the most permissive one it is headquartered in. Betting the product on the lightest rulebook is a bet you lose the moment you cross a border.
What a coordinating “floor” would actually do for you
The UN Global Dialogue is not proposing to replace the EU AI Act or overrule Bletchley. Its ambition is narrower and, for business, more useful: turn a scatter of overlapping instruments into something interoperable. The Ada Lovelace piece points to the elements worth watching, and each one maps to a commercial interest.
Shared expectations on what AI must not do would give firms a stable, universal set of red lines rather than a moving target that differs by market. Technical interoperability — common standards for transparency, evaluation, safety benchmarking and documentation — is the difference between building your compliance evidence once and rebuilding it for every jurisdiction. Shared evidence and monitoring across the AI value chain would close the visibility gap that leaves risk teams guessing.
If any of that lands, the winners are precisely the mid-sized exporters who cannot afford to run a separate compliance operation per continent.
A practical way to read this, and act on it
You do not need a seat in Geneva to respond sensibly. Treat the fragmentation as a standing feature of the market and build for it, roughly in this order:
- Map your regime exposure. List every jurisdiction each AI-enabled product actually reaches, and the binding rules in each. Most firms discover the list is longer than they assumed.
- Design to the strictest floor, not the friendliest one. Building to the toughest regime you sell into is usually cheaper than maintaining divergent versions and cheaper still than a forced retrofit.
- Instrument your evidence once, reusably. Transparency, evaluation and documentation practices are converging across instruments. Build them to a common standard now and you are ready for whichever framework hardens first.
- Track convergence, not just individual regimes. The signal worth following is not any single summit but where the instruments start to agree. That is where the durable rules will settle.
SME Advantage: A smaller UK firm can turn coherent internal governance into a sales asset this year. When a customer’s own compliance team asks how your AI is tested and documented, a single, portable answer built to the emerging floor beats a shrug — and it is exactly the answer larger, more siloed competitors struggle to give.
Four things that could still go wrong
The optimistic reading is easy. These are the failure modes worth holding in mind.
The floor becomes a ceiling. A universal minimum, negotiated to consensus, can quietly cap ambition. Once everyone agrees the baseline, the pressure to exceed it fades, and “compliant” starts to mean “adequate for the lowest common denominator.”
Consensus dilutes substance. The same breadth that gives the UN its legitimacy makes hard commitments difficult. Getting 193 states to agree tends to produce language everyone can sign precisely because it asks little of anyone.
The geopolitics does not hold. The 2024 resolutions passed by consensus, but AI is moving to the centre of military and economic strategy. As the Ada Lovelace analysis warns, absent shared governance, states optimise individually and lock into a race for dominance. Cooperation is fragile exactly where the stakes are highest.
The window closes before the work is done. Governance frameworks get harder to establish as technology and expectations entrench. The moment to shape interoperable rules is open now and will not stay open. A dialogue that talks past the window achieves nothing.
The strategic takeaway
The Ada Lovelace Institute puts the real question well: it is not whether AI will be governed, because it will be, but whether the rules get built whilst the architecture can still be shaped together. That is a diplomatic point. It is also a commercial one.
Every UK firm building or buying AI is already governed by a tangle of regimes it did not design and cannot individually change. The fragmentation is not a future risk to plan for. It is a present cost, paid in compliance overhead, blocked market access and competitive disadvantage for the firms too small to absorb it. The UN Global Dialogue may or may not deliver the coherence it is aiming for. Either way, the firms that come out ahead will be the ones that stopped waiting for a single clean rulebook and started building to the floor the rest of the world is slowly converging on.
Interoperability is not only what the UN wants from its member states. It is the smartest posture a UK business can adopt whilst the diplomats argue.
Source and attribution
This analysis draws on “The UN Dialogue of Dialogues: Building strength through complementarity” by Claire Melamed (UN Foundation) and Friederike Schüür (Ada Lovelace Institute), published 3 July 2026 ahead of the inaugural UN Global Dialogue on AI Governance in Geneva, and its references to the UN Scientific Panel on AI’s preliminary 2026 report. Original post available at adalovelaceinstitute.org.
Editorial analysis and UK business framing by Resultsense. We make sense of AI in the UK — turning research, policy and announcements into what they mean for the people building and buying these systems. For more analysis, explore our insights or get in touch.