Summits of Convenience
For one week in June, the canton of Geneva closed itself around a meeting it was not hosting. From the 11th to the 18th, the Swiss Confederation reinstated border controls while Geneva shut administrative counters, restricted stretches of motorway, and took its contraventions office offline as a precaution. Storefronts and windows were boarded up throughout the city in preparation. Not for a war, not for a flood, but because seven governments were meeting nearby.¹
Geneva was not hosting the G7. Its residents did not vote for it. Its economy absorbed measurable losses, for which a compensation fund had been assembled in advance²—an admission that the disruption was real enough to need a budget line. The summit was held forty kilometres away, across the border, in a French lakeside town, returning to the hotel it had used in 2003.³ The inconvenience was real. It was also the least interesting thing about the month. Weeks later, a few kilometres away, a different way of governing one of this century's hardest problems walked through the open door of the same city.
The Diagnostic Frame
The G7 arrives pre-framed in two ways, and both are limiting. The first frame is about security—leaders face genuine threats and the only permitted debate is about calibration. The second is about inconvenience—residents are disrupted, the optics are poor. Either way, the summit itself is treated as legitimate, necessary, and beyond examination, and all that is left to argue about is the width of the perimeter rather than what it exists to protect. Both framings leave the same question untouched: not how many checkpoints Geneva needs, but whether the institution behind them still does what it claims to do, and who gets to decide the answer.
What The G7 Claims To Do
The G7 is an informal grouping of advanced democracies that meets annually to coordinate economic policy and address whatever else is on the agenda. It was assembled in 1975, when Western economies were absorbing the oil shock and the collapse of the Bretton Woods monetary system. Fifty years on, that world has gone, and the club remains.⁴ Its present agenda spans climate, global health, digital governance, infrastructure, migration, geopolitical stability, and gender equality—a general theory of everything, which is often the sign of an institution with no single mandate in particular.
Consider the architecture. No founding treaty. No permanent secretariat of consequence. No parliament, no court, no electorate. Communiqués that bind no one.⁵ And for accountability, a mechanism of notable circularity: in 2009 the G7 established an Accountability Working Group to monitor its own commitments, publishing a comprehensive self-review every three years.⁶ The group sets the targets, drafts the language, and assesses its own performance. Calling this accountability is generous; it is a body reporting on itself, to itself, against standards it wrote itself. The difficulty of holding the G7 to account is a property of the system. Properties that survive fifty years of institutional inertia are rarely accidental, and tend to be useful to someone.⁷
The Performance Gap
Climate is the fairest test, because it is where the G7 has made the most commitments over the longest period against the most measurable targets. If the forum ought to deliver anywhere, it ought to deliver here.
It has not.
The 2025 summit was the deadline for members to phase out inefficient fossil fuel subsidies, a pledge first made in 2009.⁸ The deadline passed with G7 members still among the highest emitters on the planet and off track for their obligations this decade. At Kananaskis, the most recent summit before this one, two days of talks produced a leaders' statement that touched climate only in a passing line of the Chair's summary, while dwelling on the wildfires the same climate is producing.⁹ The pattern is by now familiar: commitments made, restated, deferred, and allowed to lapse in silence.
The word 'inefficient' has been in the pledge since 2009, and it acts much like an escape hatch.¹⁰ It is the qualifier that lets each member decide that its own subsidies are the efficient kind—a commitment built to be unfulfillable, and duly unfulfilled. This is less a failure of will than a predictable output of design. A forum with no enforcement, no independent monitoring, and no penalty for non-delivery produces exactly this. It generates commitments at scale and is under no obligation to honour them.
The Representation Problem
Suppose the G7 delivered on every promise it ever made. It would still be the wrong room. Its share of global output has fallen steadily across fifty years, as economic weight has shifted toward Asia, Africa, and Latin America.¹¹ The countries that will absorb the worst of climate breakdown, sovereign-debt distress, and extractive trade arrangements have no seat in the room and no procedure for acquiring one. The G7 has spent real credibility in the Global South by resisting reform of the IMF quota system that would shift voting power toward emerging economies¹², guarding an old distribution of influence while issuing communiqués about partnership.
Even sympathetic analysts concede the obvious: given its narrow membership and shrinking weight, the G7 cannot lead the global response to problems like pandemics or climate change, which have to be negotiated in wider and more inclusive forums.¹³ The G7 persists as an informal steering committee for an order it authored, for purposes it defined, in a century that has ended. Its authority rests less on mandate than on incumbency.
The Missing Principals
Here the usual critique stops too early. The standard complaint is that the leaders are inadequate, captured, or insincere. The deeper issue is that they are, increasingly, not the ones deciding. Heads of state are no longer the de facto rulers of the global economy, if they ever were. The conditions that shape lives—capital allocation, trade flows, supply-chain geography, energy infrastructure, the price of money—are set largely by actors absent from the Évian guest list: asset managers moving sums that dwarf national budgets, sovereign wealth funds, corporate lobbying coalitions, and the revolving door between private finance and the public bodies meant to regulate it.
The leaders know this. The genuinely consequential business of these summits happens in the bilateral meetings and the unminuted dinners, much of it devoted to managing the volatility of one or two unpredictable actors rather than to governing much at all. The seven principals are, in large part, managing the consequences of decisions taken in rooms they were not in. The spectacle has a function. The motorcades, the family photograph, the choreographed communiqué locate accountability in seven recognisable faces and make global power feel personal and legible. While the public weighs whether these seven did well, the harder question—who governs the world economy, and by what right—goes unasked. The photograph is where attention goes and yet, the governing happens elsewhere.
The Security Theatre Loop
The perimeter around Évian is Genoa, 2001: a protester killed, hundreds injured, and police conduct later ruled by the European Court of Human Rights to have amounted to torture.¹⁴ The lesson drawn was not to reconsider a format that placed leaders and citizens on a collision course, but to enlarge the perimeter. From this grew a doctrine of closed logic: if disorder breaks out, the perimeter was too small; if order holds, the perimeter worked. No outcome argues for less.
Each summit inherits the last one's footprint as its floor and builds upward. The costs fall on residents who had no say in the decision and will have none in the review. A vast, visible security operation also does quieter work: it reframes dissent as a matter to be managed rather than heard, and makes peaceful protest, historically the means by which citizens have moved institutions, logistically and legally harder. The perimeter keeps bodies out, and narrows the conditions under which objection is possible.
The New Domain
The G7 has extended its governance claims to frontier AI. The Hiroshima AI Process, launched at the 2023 summit under Japan's presidency and carried forward under Italy's, produced a Comprehensive Policy Framework: twelve International Guiding Principles for all AI actors, and an eleven-action Code of Conduct for the organisations developing the most advanced systems.¹⁵ The documents are serious and carefully drafted. In their accountability architecture, they are also familiar : voluntary and self-reported.
Apply the same questions. Who is absent? The governments of China, India, and most of the world beyond the G7, much of the territory where AI development and deployment will actually occur. What is the enforcement mechanism? There is none. Who monitors compliance? At the G7's request, the OECD built a voluntary Reporting Framework through which companies describe how their practices align with the Code; some of the largest developers took part in the pilot.¹⁶ Notably, the pilot submissions were not disclosed, a transparency instrument whose first responses were kept confidential to encourage candour.
The firms shaping the trajectory of frontier AI are present through the B7 business group, which files its recommendations and is gone by the time decisions are made. The civil-society organisations representing the people who will live with automated decision-making and AI-enabled surveillance submit a parallel document, with a fraction of the access and none of the leverage.¹⁷ Both are heard. They are not heard equally.
The Hiroshima Process applies the G7's governance model to the most consequential technology of the decade. The same forum that has spent seventeen years not phasing out fossil fuel subsidies is now custodian of voluntary principles for systems its members are still working to define. The escape hatch is built into the architecture, as it usually is. The G7 has positioned itself as a room where global AI governance happens. It is worth asking what was being built, that same month, in the larger room a few kilometres away.
A Few Kilometres Away
On the 6th and 7th of July, for the first time, all 193 UN member states convened in a single forum dedicated to governing artificial intelligence: the inaugural Global Dialogue on AI Governance, mandated by the General Assembly and co-chaired by the ambassadors of El Salvador and Estonia. Where Évian required a perimeter, this required a registration desk. Developing countries and the Global South took part with full standing to shape the outcome, not as observers.¹⁸
The Dialogue did not appear from nowhere. It is a product of the Global Digital Compact, adopted in September 2024 as part of the Pact for the Future, the first universal intergovernmental agreement on digital and AI governance, whose terms of reference the General Assembly adopted by consensus in August 2025, after a year of negotiation co-facilitated by Costa Rica and Spain.¹⁹
Five days before the Dialogue, its scientific counterpart reported. The Independent International Scientific Panel on AI—forty experts drawn from all five UN regions, serving in their personal capacity, co-chaired by Yoshua Bengio and Maria Ressa—released its first assessment on 1 July.²⁰ Its message to governments was not consoling: current safeguards cannot keep pace with AI's capabilities, and, Bengio told diplomats, science cannot yet guarantee that advanced systems will not cause catastrophic harm. Ressa reduced the governance problem to a sentence: the world cannot govern what it cannot understand.²¹
The Panel is, by deliberate design, non-prescriptive: it supplies evidence rather than policy, precisely so that political contestation cannot compromise the science.²² This is a different thing from the G7's voluntariness, and the distinction matters. The Hiroshima Code is unenforceable because it asks the governed to police themselves. The Panel is non-prescriptive because it divides the labour: science establishes a shared baseline, and the Dialogue, where every state has a voice, is where prescription is meant to be argued out. The design reflects hard trade-offs rather than evasion of them: an independent scientific core to protect rigour, a state-inclusive layer to carry political concerns, non-state observers without a vote, in the manner of the IPCC or the ILO's tripartite model.²³ Because AI outpaces the five-to-eight-year cycles of climate assessment, the Panel is built for speed: an annual report with interim updates, rather than a once-a-decade verdict.²⁴
None of this should be oversold, and honesty here is the point. The Dialogue is not binding; a forum without enforcement can still end in a document nobody must honour; the very failure diagnosed above. Its year of consultations drew more than 1,500 written submissions and exposed the fault lines plainly: governments alone ranked capacity-building first, while most other constituencies ranked safety first.²⁵ And the Panel's central warning named the deeper problem this essay began with: the concentration of consequential decisions in very few hands, with two states now commanding the overwhelming majority of the world's advanced computing power.²⁶ Inclusion on this scale is a precondition for legitimate governance. It is not yet governance.
The Question Worth Keeping Open
Set the two meetings side by side. In June, seven governments met behind a cordon and asked to be judged on their own account of themselves. In July, near the same corner of the same lake, 193 met in the open and agreed, at minimum, on what they do not yet know. Neither has binding force. But only one is built on the premise that the people affected should have a seat.
The lessons the G7 might take are not mysterious, and have sat in policy journals for years: bind commitments to independent monitoring rather than self-assessment; consult the African Union, ASEAN, and other regional bodies as a matter of structure rather than courtesy; disclose the interests present in the margins.²⁷ The G20 itself was created in 1999 because a narrower club could not manage a crisis alone.²⁸ These formats widen when the will exists.
Strip away the security-versus-freedom framing, which argues on the institution's own terms, and a plainer account remains. Citizens are asked to absorb real costs—economic, civic, political—for forms of governance built for a world that has changed beyond recognition, and too often closed to the people who live with the results. The better design is no longer hypothetical: it held its first session in July, a few kilometres from the perimeter. Whether it succeeds is undecided. Whether it points in the right direction is not.
There is a third option, between romanticising the summit and dismissing cooperation as theatre: to ask of every forum that shapes our lives what it is for and whom it serves—and to keep asking. A perimeter can be drawn around a town for a week or a weekend, but it cannot be drawn around a question.
Path:OS works at the intersection of governance, systems change, and civic capacity. This piece reflects the analytical perspective of the practice. Empirical claims are sourced; where the evidence thins, the text says so.
Notes & Sources
Republic and Canton of Geneva, "Sommet du G7 – Évian 2026." ge.ch/dossier/sommet-du-g7-evian-2026. Swiss Federal Department of Foreign Affairs (FDFA), "G7 Summit in Évian." eda.admin.ch/en/g7-summit-in-evian
Republic and Canton of Geneva, "Sommet du G7 – Évian 2026." ge.ch/dossier/sommet-du-g7-evian-2026
Élysée, "2026 G7 Summit of Évian." elysee.fr/en/G7evian. Council of the European Union, "G7 summit, Évian, 15–17 June 2026." consilium.europa.eu
Council on Foreign Relations, "What Does the G7 Do?" cfr.org/backgrounders/what-does-g7-do. Climate Diplomacy, "G7 Summit 2026." climate-diplomacy.org/events/g7-summit-2026
Climate Diplomacy, "G7 Summit 2026" (informal body: no legal existence, no permanent secretariat, no ex officio member). climate-diplomacy.org. Banque de France, "G7 Évian 2026." banque-france.fr/en/g7-evian-2026
Global Affairs Canada, "G7 Canada 2025 Comprehensive Accountability Report." international.canada.ca/en/global-affairs/corporate/reports/g7-2025-accountability-report
German Institute of Development and Sustainability (IDOS), "How the G7 reviews its work on development: a case study of internal accountability." idos-research.de
International Institute for Sustainable Development (IISD), "Four Ways the G7 Can Show Leadership on Fossil Fuel Phase-Out" (subsidy pledge made since 2009; 2025 deadline set in 2016). iisd.org/articles/insight/g7-leadership-fossil-fuel-phase-out
IISD, "Inside the G7: Environment and energy" (G7 members among highest emitters; Kananaskis sidelined climate while emphasising wildfires). iisd.org/inside-g7-environment-energy
IISD, "Four Ways the G7 Can Show Leadership on Fossil Fuel Phase-Out" (the "inefficient" qualifier, in the pledge since 2009). iisd.org/articles/insight/g7-leadership-fossil-fuel-phase-out
Council on Foreign Relations, "What Does the G7 Do?" (declining share of global output; questioned relevance). cfr.org/backgrounders/what-does-g7-do
ENSURED, "Relic of the Past or Catalyst for Change? The G7 and the Future of Global Governance" (G7 resistance to IMF quota reform; standing in the Global South). ensuredeurope.eu/publications/g7-and-global-governance
ENSURED, "Relic of the Past or Catalyst for Change?" (the G7 cannot lead on pandemics or climate without wider, more inclusive forums). ensuredeurope.eu
G7 Research Group, University of Toronto, "2001 Genoa Summit Assessment" (death of Carlo Giuliani; ~230 injuries). g7.utoronto.ca/evaluations/2001genoa/assess_summit_hajnal.html. The Irish Times, "Italy found guilty of using torture during G8 protest" (European Court of Human Rights, Diaz raid, 7 Apr 2015). irishtimes.com
On the Hiroshima AI Process framework—twelve International Guiding Principles for all AI actors and an eleven-action Code of Conduct for developers of advanced systems: CSIS, "Shaping Global AI Governance" (csis.org); OECD, "Hiroshima AI Process" (oecd.org / transparency.oecd.ai).
OECD, "OECD launches global framework to monitor application of the G7 Hiroshima AI Code of Conduct" (voluntary, self-reported Reporting Framework; largest developers contributed to the pilot; Feb 2025). oecd.org. OECD, "Towards a Hiroshima AI Process Code of Conduct Reporting Framework" (pilot-phase submissions not disclosed). wp.oecd.ai
U.S. Chamber of Commerce, "What is the B7?" uschamber.com. Civil7, "What is C7?" (civil-society engagement group; constituencies including the Global South). civil7.org/about-c7/
First Global Dialogue on AI Governance, Geneva, 6–7 July 2026: first convening of all 193 UN Member States on AI governance; co-chaired by Amb. Egriselda López (El Salvador) and Amb. Rein Tammsaar (Estonia); opened by the UN Secretary-General; Global South participating with full standing. UNESCO, "UN Global Dialogue opens with urgent call for safe and inclusive AI" (unesco.org); UN News, news.un.org/en/story/2026/07/1167873.
Global Digital Compact (UNGA resolution A/RES/79/1), adopted 22 September 2024 within the Pact for the Future; terms of reference for the Panel and Dialogue adopted by consensus in A/RES/79/325, 26 August 2025, co-facilitated by Costa Rica and Spain. un.org; dig.watch/processes/global-digital-compact
Independent International Scientific Panel on AI, "Preliminary Report" (New York, 1 July 2026): first global independent scientific assessment of AI opportunities, risks and impacts; 40 members from all five UN regions serving in personal capacity; co-chaired by Yoshua Bengio and Maria Ressa; comprehensive report due 2027. un.org/independent-international-scientific-panel-ai
On the Panel's central findings—that safeguards cannot keep pace with capabilities, and that science cannot presently guarantee against catastrophic harm (Bengio), and Ressa's framing that governance requires understanding: UN News (news.un.org); Tech Times, "UN AI Governance Summit" (techtimes.com); AI Weekly (aiweekly.co).
On the Panel's deliberately non-prescriptive mandate (evidence, not policy recommendations, to avoid politicising the science): Inter Press Service, "UN Artificial Intelligence Panel Launches Report" (ipsnews.net); alignmt.ai analysis.
On institutional-design trade-offs—inclusion versus decision-making capacity, models such as the IPCC and the ILO tripartite system, and non-state observers without voting rights: Simon Institute for Longterm Governance, "Blueprints: Design Options" and "Recommendations for the Independent International Scientific Panel on AI and the Global Dialogue on AI Governance." simoninstitute.ch
On the pacing problem and "speed by design" (annual assessment with interim updates, versus the five-to-eight-year IPCC cycle): Simon Institute, "Three considerations for the Independent International Scientific Panel on AI" and "Adoption of the Panel and the Global Dialogue." simoninstitute.ch
On the non-binding character of the Dialogue and the consultation results—more than 1,500 written submissions, with governments alone ranking capacity-building first and most other groups ranking safety first: UNESCO (unesco.org); ThePlanetTools.ai analysis.
On the concentration of AI decision-making power and compute—the Panel's framing that a small number of companies and states hold outsized influence, and that roughly two states command the majority of advanced compute: Tech Times (techtimes.com); Bösl, "UN Independent Scientific Panel AI—Report Analysis" (boesl.org).
Center for Strategic and International Studies (CSIS), "A Reimagined G7" (formal consultation with the African Union, ASEAN and other regional bodies). csis.org/analysis/reimagined-g7
Brookings Institution, "Pragmatic Reform of Global Governance: Creating an L20 Summit Forum" (the G20 created in 1999, after the narrower club proved insufficient for the 1997–98 financial crises). brookings.edu