Global talks agree principles. Who writes the rules?
On 3 September 2026, the day after G20 ministers closed a technology meeting in North Carolina with a consensus statement, the Lowy Institute’s Interpreter published a short essay by Sameera Pillai, a Canberra-based policy and research officer, under the summary line “Superpower rivalry has stalled global AI rules, but regional blocs like ASEAN are moving fast enough to make rules stick.” Her case is written mainly about the Indo-Pacific and never mentions the UK. Read from London seven weeks later, it describes Britain’s position with uncomfortable precision: the UK was in the room when G20 ministers reached a light-touch consensus, it has chosen not to adopt an AI Act of the kind Pillai cites as proof that a smaller group can legislate faster than global consensus, and it still has no AI legislation of its own.
Strategic Insight: Pillai’s “third path” is a route through regional coalitions, so that countries can shape AI standards “without having to fully align with either superpower”. It is a path that runs through a bloc. A middle position that belongs to no bloc is a different thing, and it is not yet clear which of the two the UK holds.
What does the Lowy essay actually argue?
Pillai starts from a split. In her account the US pursues a security-focused approach built on “trusted networks” of allies, whilst China backs multilateral arrangements steered by the state, such as the World AI Cooperation Organisation. She engages with Dario Amodei’s call for a coalition of democracies, which she finds strong in logic but unclear in scope: what counts as a democracy now, and which side of the line Türkiye or India would fall on.
Her alternative is regional. ASEAN and the Pacific Islands Forum, she writes, already have momentum on AI rules, which gives them a head start over any brand-new coalition. Global processes, the AI Safety Summits among them, produced communiqués that were “valuable but non-binding”, whereas the EU managed to turn its AI Act into binding law in a matter of years. She concludes that “fewer, more aligned members can legislate and enforce faster than global consensus allows.”
The real story is in the mechanism, not the map
The strongest part of the essay is its account of why a regional rule sticks. Pillai’s answer is trade. Where access to a market depends on compliance, alignment stops being voluntary. She points to the Singapore-Australia Digital Economy Agreement, which already treats AI as a field of formal cooperation next to binding commitments on data flows, and suggests the same structure could be extended to AI governance inside wider agreements, naming the trans-Pacific trade pact known as the CPTPP.
That detail matters for Britain more than the essay lets on. Britain’s CPTPP membership took effect on 15 December 2024, making it the first member that was not among the original signatories. The trade bloc Pillai proposes as a carrier for binding AI rules is one the UK already belongs to.
| Marker | Status | What it shows |
|---|---|---|
| G20 Innovation Ministerial, Chapel Hill | Consensus reached on the Carolina Principles; the EU and UK listed among those present | The global layer agreed shared principles |
| EU high-risk rules for stand-alone systems | Now take effect on 2 December 2027, under Regulation (EU) 2026/1744, which is in force | The EU bloc keeps binding law but has moved its own start date |
| UK AI-specific legislation | None, according to the House of Commons Library on 14 October 2026 | The UK has no statute of its own to bring to any bloc |
| UK membership of the CPTPP | In force since 15 December 2024 | Britain already sits in the trade bloc Pillai names as a template |
Strategic Reality: Pillai’s evidence for regional speed is the EU AI Act. The EU’s own 2026 amending regulation now defers the Act’s rules for stand-alone high-risk systems from 2 August 2026 to 2 December 2027. Writing binding law quickly and applying it quickly turn out to be separate achievements.
Three events, one pattern?
The brief case for treating September’s G20 meeting, the EU’s July amendments and the UK’s absent bill as one story is that each shows the same division of labour: principles at the global level, binding rules (or their absence) at the level of a bloc or a single state.
The G20 agreed to keep its hands light
Reuters reported from Chapel Hill on 1 September that Washington was urging G20 governments to stay hands-off on AI and not to write fresh rules for it. White House technology adviser Michael Kratsios told officials that countries signing the Carolina Principles agree to “reserve new regulation for novel considerations”, and a White House official added that Washington would argue against setting up new AI regulators. The White House’s own readout says ministers “reached consensus on the Carolina Principles” and lists both the European Union and the United Kingdom among those present. We covered the meeting as it happened.
This fits Pillai’s pattern rather than breaking it. Kratsios summed up the meeting as recognising that “flexible policy frameworks crafted to promote innovation will drive economic growth and prosperity”. What the statement contains, by the White House’s own description, is shared principles.
Critical Context: The EU was in the room for that consensus. A bloc can join a light-touch global statement and keep a binding rulebook at home, which fits the line Pillai draws between non-binding global communiqués and binding law made by fewer members. The question for the UK is what sits at its own second level.
The EU kept its rulebook and moved the clock
Regulation (EU) 2026/1744, the Digital Omnibus on AI, is dated 8 July 2026, appeared in the Official Journal on 24 July and is listed on EUR-Lex as in force. It sets the AI Act’s high-risk chapter to take effect on 2 December 2027 for systems classed as high-risk under Annex III, and from 2 August 2028 for those under Annex I, the AI built into regulated products. When the political deal was struck in May, Pinsent Masons reported that the delay was meant to give EU policymakers more time to produce standards, tools and guidelines, and that the wider omnibus package aimed to simplify rules for the sake of EU competitiveness. We analysed the omnibus and what it meant for the UK in June.
The EU bloc has not given up binding rules. It has kept them and deferred the date they bite. Speaking in May, Pinsent’s Nils Rauer warned that the changes leave companies “reliant on secondary measures that are still to be finalised”.
Britain made a promise and has not legislated
The Commons Library’s briefing of 14 October is blunt: “The UK does not have any AI-specific regulation or legislation covering AI as a technology.” It records that Labour’s 2024 manifesto and the 2024 King’s Speech signalled “binding regulation on the handful of companies developing the most powerful AI models”, and adds: “Legislation, however, has not yet been forthcoming.” The IAPP noted that the King’s Speech of 13 May 2026 contained no standalone AI bill at all. In February 2025, as the Library records, the government said “most AI systems should be regulated at the point of use”.
That same King’s Speech, according to the IAPP, carried bills on alignment with the EU. Whether any of that alignment reaches AI rules could not be confirmed from the material available to us; the relevant section of the IAPP analysis is behind its membership wall.
What makes a regional rule stick, on Pillai’s account:
- Trade conditionality: rules become binding when market access depends on them, which is why she looks to digital economy agreements and the CPTPP.
- Fewer, closer members: a smaller, like-minded group can pass and police rules more quickly than a global forum.
- Risks specific to the region: she cites electoral deepfakes in Southeast Asia, and the separate Grok bans imposed by Malaysia, Indonesia and the Philippines over deepfakes made with it.
- A treaty to carry it: ASEAN’s Digital Economy Framework Agreement, which she describes as treaty-level, “will embed AI governance provisions”.
None of those four conditions currently describes a UK AI rule, because there is no UK AI rule for them to describe.
Who feels the difference?
Pillai flags the cost of a world of blocs herself: differing standards across regions “could create compliance hurdles for AI developers operating across borders”. For British organisations, the split shows up in different places depending on where they sell and what they build.
| Group | What changes for them | What is still unknown |
|---|---|---|
| Businesses with duties under the EU high-risk regime | Stand-alone high-risk rules start on 2 December 2027; product-embedded AI follows on 2 August 2028 | Rauer said in May that firms would depend on secondary measures not yet finalised |
| Developers of the most powerful models working in the UK | No binding UK regime yet, despite the 2024 promise | Whether and when the promised legislation arrives |
| The UK government, as G20 president in 2027 | Responding to a question about the G20 presidency, AI minister Kanishka Narayan said AI would top the agenda, the Herald Series reported, and that the UK wants to lead | Whether the UK’s G20 year aims for anything beyond principles |
| UK traders inside the CPTPP | The bloc Pillai names as a possible carrier for binding AI provisions already covers them | Whether binding AI provisions of the kind Pillai describes ever reach the CPTPP |
Hidden Cost: Pillai’s suggested routes to harmonisation are common technical benchmarks resting on the OECD’s AI principles, or standards designed to work together. In our view, those work best between parties that each bring a rulebook to the table, and the UK currently has no AI statute to bring.
What is unresolved for a country between blocs?
Is the middle position a policy or a vacancy?
Pillai’s third path gives countries agency through regional coalitions, and she argues that “some form of binding governance is better than none at all”. The UK’s current arrangement, a global light-touch statement plus sector regulators and no AI statute, keeps options open without exercising any of them. That can be a deliberate strategy. It can also be what having no bloc looks like from the inside. The evidence so far does not settle which.
Would a bloc want Britain on Britain’s terms?
Pillai concedes that regional bodies are not immune to power politics: “A dominant economy may hold outsized influence over which standards are set.” On that logic, a UK that joins an AI arrangement late, in the EU’s orbit or the CPTPP’s, would be joining standards others had already shaped. That is a cost of waiting, and it grows as each bloc’s rules harden.
Does deferral change the EU’s pull?
Pillai calls the EU AI Act “the world’s first comprehensive, binding AI law”, but its high-risk rules now start more than a year later than first planned. For a UK weighing whether to align, the bloc’s rulebook is now both settled in law and further off.
Reality Check: Pillai does not claim regional coalitions will end the US-China contest. Her words: they “will not dissolve the binary race to govern AI between the US and China”. Her claim is that they offer “a more realistic third path”.
What to watch
Four checkable events will show whether Britain picks a bloc, builds one, or stays between them.
- December’s Miami summit of G20 leaders. If the leaders’ text repeats the Carolina language, the global layer is settled as principles only, and binding rules are left to blocs and states, which is the pattern Pillai describes.
- The UK’s G20 presidency in 2027. Narayan has said AI will top the agenda, according to the Herald Series. A presidency that pursues principles would fit the UK’s current position. One that pursues an agreement with obligations, or an AI chapter in a trade framework, would be the UK choosing a bloc. A Liberal Democrat MP has already pressed the minister on what that agenda should contain.
- ASEAN’s Digital Economy Framework Agreement. Pillai says it will carry AI governance provisions. If it does, her thesis gets a treaty-level test in the region she writes about.
- The promised UK legislation. The Commons Library records the 2024 promise of binding rules for frontier developers as unfulfilled. A bill would give the UK something to bring to any bloc. Another session without one would suggest Britain’s version of the third path is, for now, a description of where it happens to be.
Pillai ends on Amodei’s likening of AI’s geopolitical weight to that of nuclear weapons, and on the difference she sees: much of the arms control architecture was built after the Second World War, whereas with AI, “the window to get governance right is still open”. For the UK, the narrower question is whether it wants a seat in the coalitions writing binding rules while that window lasts, or whether it is content to sign the global statements and wait. Our earlier piece on where the UK’s AI strategy sits among national governance models set out the options.
Source: The quiet regionalisation of AI governance (The Interpreter, Lowy Institute, 3 September 2026), by Sameera Pillai. Additional sources: US urges hands-off approach to AI regulation at G20 tech meeting (Reuters, via Geo News, 1 September 2026); G20 Innovation Ministerial Concludes with Consensus Statement (The White House, 2 September 2026); Regulation (EU) 2026/1744 (EUR-Lex); Rules on ‘high-risk’ AI to be delayed under EU ‘omnibus’ deal (Pinsent Masons Out-Law, 7 May 2026); AI regulation in the UK (House of Commons Library, 14 October 2026); King’s Speech signals diffuse UK digital policy agenda, but no AI bill (IAPP, May 2026); MP urges global action to prevent superintelligent AI (Herald Series, October 2026); UK joins CPTPP: implications for businesses (EY, 16 December 2024).
This strategic analysis was written by Resultsense, a UK-focused AI news and analysis publication. We will be watching whether Britain’s G20 presidency in 2027 carries any binding element, and whether a UK AI bill finally reaches Parliament. Read more analysis at Insights, or get in touch.