Loading the longlist of options
Each column is one option, numbered as in the longlist. The upper chart shows how achievable it looks at a 2027 G20. The rows below show likely positions: the US, China, the EU and the UK, then France, Germany and Italy only where they part company with the EU, then India, and Russia, Saudi Arabia and Indonesia (RU, SA, ID) as one group. The last row flags options where other members are likely to cause trouble. Hover an option for its summary, or any dot for the reasoning behind that position; click to jump to the option. On a small screen, scroll sideways.
Everything is on the table here, from restating what exists to new rules outside consensus. Each option has political strengths and weaknesses to start from. Open "Detail" for the precedent, delivery body, positions and the main risk.
Ordering. Options run from least to most ambitious. Ambition means how far an option moves beyond what the G20, or the US, has already agreed. The lower bands lean on precedent; the higher bands break new ground, ask more of members and have fewer existing texts to point to.
Scores and positions. Ambition, achievability and every country position are judgements drawn from October 2026 research on G20 outcomes since 2008 and on what Trump administrations sign. They are there to start a discussion, not settle it.
US view. Based on what the administration has signed or drafted (Osaka 2019, Pax Silica, New Delhi 2026, Chapel Hill 2026) and what it has rejected by name (Paris 2025, "global governance" of AI at the UN).
Europe and the UK. The UK row shows the likely position of the host government. The EU row reflects the Commission and the broad EU line. France, Germany and Italy are shown separately only where their governments' recent stances point elsewhere. Those splits rest on thinner evidence than the US and China positions.
Not mutually exclusive. A realistic package combines several options: a short consensus text, delegated work in band 3, and a welcomed annex from band 5.
Checks needed. Some 2026 points rest on single or lower-quality sources: details of Évian, the substance of the US–China AI dialogue, and the UK's reported presidency priorities. The Miami summit (14–15 December) may shift the baseline.
Major AI policy updates from G20 members, the EU and the African Union, researched once a day by a scheduled task. Each story links to its source, preferring official texts and wire services to aggregators and blogs, and the list aims to cover Europe, Asia and the Global South as well as the US and UK. Most summaries end with a sentence giving our own reading of what the story means for the G20; from 27 October 2026 that sentence starts "Dashboard view:". When a story materially shifts a government's position on an option, or raises a genuinely new idea, the same task updates the dashboard and logs the change here.
Pick options to form a package and see who would likely sign, who needs work and who would block. The US and China carry extra weight: either one blocks if it is unlikely to accept any option in the package. Other governments object to an option they are unlikely to accept, which can usually be handled with softer wording, a trade or a chair’s summary. They block if two or more of them object to the same option, or if one is unlikely to accept a third or more of the package. Read the result as a guide to where effort is needed; it does not predict the final text. Start from a preset or build your own. Positions are the same judgements used across the dashboard.
How open the climate for AI is in each G20 member, from public attitudes and how light the rules are.
Public optimism (across) against how restrictive its AI rules are (up). Shaded areas group countries with similar publics and rules. Countries in grey; coloured dots mark the US, China and the average of each bloc. Members without a survey since 2024 are left out.
Bloc averages in colour, G20 average in lime, every member in grey. Focus on a bloc or find a country above to fade the rest.
Survey waves for the backbone measure. G20 average in lime.
Not in the index, but in the data library below: IMF AI preparedness, trust in government, AI investment, notable models, AI laws passed, supercomputers and international initiatives.
Ranked by index. Each part is scaled so the lowest G20 member scores 0 and the highest 100; the raw figures behind each part sit beside it. ⚑ means one or more opinion measures are missing: hover it to see which. Click a heading to sort.
Data years: optimism 2026 (Ipsos AI Monitor), AI use and trust in AI 2025 (KPMG and University of Melbourne), rules 2026 (this dashboard).
| Part 1 · Public attitudes · 50% | Part 2 · Policy openness · 50% | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| # | 0–100 | scaled 0–100 | Ipsos % | KPMG % | KPMG % | scaled 0–100 | restrictiveness 1–5 | ||
The international AI initiatives each G20 member has joined, oldest first. Click a heading to sort by it; hover a heading for what it is.
Compare G20 members across measures of public opinion, policy, the AI economy and international engagement. Pick one measure to rank members, or two to plot against each other.
What each government is likely to refuse, what it wants, and the options most at risk from it, with a record of its main AI laws and regulations. Option numbers link to the longlist. These are judgements from public positions as of October 2026. Each cohort shows its bloc memberships, a confidence rating and its sources. Ratings are lower where a government has said little in public about G20 AI, as with Mexico, Türkiye and the African Union.
AI laws and rules of G20 members that start to apply between now and the end of 2028.
The lesson of Chapel Hill is that AI substance gets agreed at ministerial level and leaders endorse it. This is a proposed sequence. The UK has not yet published its 2027 calendar, so timings are indicative. Option numbers link to the longlist.
What agreed texts at international summits have focused on since 2023. Each column is a summit; each coloured band is a theme, as thick as its weight in the final text. Point at a band for what the text said. Pick a country to see what it asked for and whether it landed.
Reading the chart. Band thickness: 0 absent, 1 passing mention, 2 a substantive commitment, 3 a central focus. Country markers: filled means the ask landed in the text, half means partly, hollow means it did not. The label under each column is the format. Scroll sideways on a small screen.
Weights. Each summit’s agreed outcome text was read against twelve fixed themes and given 0 to 3 for prominence. These are judgements from the text, checked against official readouts, and are comparable across summits only roughly: a long G7 communiqué and a two-page declaration both top out at 3.
Asks. What a government pushed for going in, taken from its own statements, briefings and reputable reporting, and whether it made the final text. Only sourced asks are shown, so a missing marker means no evidence was found, not that a country asked for nothing. Hosts are over-represented because their aims are best documented.
Caveats. G7 Evian 2026 had no stand-alone AI text, so its weights draw on a ministerial text and readouts. The US–China row is a readout of a dialogue, not a negotiated text. The Council of Europe treaty is included as the only legally binding text in the period.
Each family’s share of all theme weight in that year’s agreed texts.
They sign more than their rhetoric suggests, across both terms. The texts share a recognisable shape.
Both signed the New Delhi Declaration (February 2026) and joined consensus at Chapel Hill (September 2026), so those texts are the safest place to borrow wording. Quote them exactly. Chapel Hill says the benefits of innovation should be "widely shared"; New Delhi says they should be "shared by humanity". Neither uses "equitably shared". New Delhi mentions "open-source AI applications"; Chapel Hill does not mention open source. China's preferred terms come from its 2023 Global AI Governance Initiative and its July 2025 Global AI Governance Action Plan. It reads "trusted" and "like-minded" framing as aimed at itself, and it wants the UN at the centre of AI governance.
| Date | Instrument | Format | Binding? | Key features |
|---|
Real, but transactional and short-lived. Deals are narrow swaps agreed at leader level, time-limited, and recorded in separate national readouts that often differ. The most relevant to AI are chip-sale decisions: Trump allowed Nvidia H200 sales to approved Chinese buyers in December 2025, but Beijing told firms not to buy and few chips had shipped by mid-2026. In September 2026 the two agreed their first AI channel under Trump II: a dialogue (the US fact sheet calls it "Super Intelligence") and an incident channel. Its next exchange is due by November 2026, so it has yet to be tested, and no commitments are attached.
| Date | Agreement | Content | How durable |
|---|
Yes. A G20 text gives wide legitimacy but only lowest-common-denominator content. A US–China channel is where frontier risk is discussed, but it is fragile and untested. A smaller group can carry ambition, but today's clubs are rival blocs: Pax Silica and China's WAICO. The US says WAICO members cannot be trusted Pax Silica partners. The UN track is the venue China prefers and the US distrusts. The G20 already has tools for combining these.
The UN track. The UN set up an Independent International Scientific Panel on AI and a Global Dialogue on AI Governance by consensus in August 2025. The Dialogue first met in Geneva on 6–7 July 2026 and meets next in New York, planned for May 2027, six months before Manchester. The Manchester text should note the Panel and the Dialogue in neutral terms and present G20 work as practical follow-through. A text that competes with the UN would lose China and much of the Global South; one that crowns the UN as the centre would lose the US.
| Dimension | G20 consensus text | US–China bilateral | Smaller group of G20 members | UN track |
|---|