Source-linked AI summary
Architecture Without an Architect? Global Governance of Artificial Intelligence in a Divided World
Simon Chesterman
TL;DR
Global AI governance faces fragmented and unevenly representative responses to a fast-moving technology, while the distribution of power complicates collective action. The review uses Maas's lenses of sociotechnical change, governance disruption, and regime complexity to assess existing efforts. It concludes that governance is already emerging, but public authority and private control over frontier AI do not align neatly.
Problem
Global AI governance must address structural challenges such as red lines for uncontrollable AI and transparency, while power over consequential frontier-AI decisions is divided between states and private firms.
Source: §2 · p. 8, §3 · p. 12Method
The review examines Maas's framework through sociotechnical change, governance disruption, regime complexity, and his proposed toolkit for choices about advanced AI.
Source: §1 · pp. 5–6Results
The review concludes that global AI governance is already an emerging, fragmented architecture shaped by interactions among technology, law, and institutional choice.
Source: §3 · p. 13Takeaways & Limitations
States retain market and strategic-input powers, but many consequential frontier-AI decisions are made inside private firms, complicating efforts to build on the existing architecture.
Source: §3 · pp. 11, 13The review acknowledges it may share omissions in its treatment of perspectives beyond the Global North.
Source: §3 · p. 11
Abstract
from arXiv · showhide
Artificial intelligence presents an unusually difficult problem for global governance. The technology develops rapidly, crosses borders easily, and is shaped by actors whose resources and capabilities may rival those of states. Yet international responses remain fragmented, unevenly representative, and overwhelmingly non-binding. The challenge is therefore not simply to identify appropriate rules or institutions, but to understand who has the capacity and incentive to create, enforce, and adapt them. This review essay examines these questions through Matthijs Maas's Architectures of Global AI Governance. Maas offers an ambitious framework for thinking about AI governance through the lenses of sociotechnical change, governance disruption, and regime complexity. His account usefully resists both technological determinism and the search for a single institutional blueprint, emphasizing instead the possibilities of a fragmented and evolving governance architecture. The essay argues, however, that institutional design cannot be separated from the distribution of power. Maas frequently invokes what "we" should do about AI, but that collective subject obscures important differences among states, international institutions, and technology companies. States retain formidable powers over markets, infrastructure, strategic inputs, and firms themselves. At the same time, many consequential decisions about frontier AI - what is built, how quickly, with what safeguards, and when it is released - are concentrated within a small number of private companies. The central problem of global AI governance may therefore be less architecture without an architect than an emerging architecture shaped by multiple actors possessing different forms of power, divergent incentives, and no common set of plans.
1 Lenses
Maas uses sociotechnical change, governance disruption, and regime complexity to analyze AI governance without assuming either technological inevitability or a single institutional blueprint. The review values these lenses but stresses that evolving governance also entails difficult choices about timing and institutional legitimacy.
1 Lenses
- Sociotechnical change helps regulators allocate scarce attention and weigh upstream intervention against downstream action under uncertainty.Early action may be easier but uncertain, while delay can clarify risks and increase the costs of addressing them.Source: §1 · p. 5
- AI mostly transforms activities within traditional jurisdictions, creating practical challenges for existing norms without requiring a wholesale governance overhaul.The review cites autonomous vehicles and vessels as cases where legal rules may need amendment, while warning against overextending regulatory responses.Source: §1 · pp. 5–6
- Regime complexity describes overlapping global governance mechanisms and offers scope for experimentation and institutional learning, but its optimistic account meets a legitimacy problem.The review notes that international law's legitimacy may be at its lowest ebb.Source: §1 · p. 6
- Maas's framework focuses on AI capabilities, including whether applications are regulation-tolerant or regulation-resistant, while recognizing that informal norm-setting takes time.The review notes that smaller, like-minded groups may settle norms even as this fragments the regime complex, and that the wider public may not have time to wait.Source: §1 · pp. 6–7
2 Action and Reaction
The review traces AI governance across military, conventional, and advanced AI, finding reactive, uneven efforts alongside a deeper problem: who has the power and incentive to shape global rules? It argues that calls for collective action cannot by themselves resolve gaps in participation, coordination, or the allocation of authority.
2 Action and Reaction
- Major institutional change often requires crisis, while AI poses a fast-moving challenge with no consensus and even less willingness to act in a coordinated manner; the 2025 meeting favored promotion over safety.The 2023 Bletchley Park AI Safety Summit launched safety institutes across multiple continents; by 2025 the meeting had abandoned AI Safety in favour of ‘AI Action’.Source: §2 · pp. 7–8
- Governments tolerated Ukraine's use of lethal autonomous weapons amid sympathy for its cause, while Kosovo intervention arguments were later cited by Russia over Crimea and its 2022 invasion.The review uses these cases to illustrate inconsistent responses to contested uses of force and autonomy.Source: §2 · pp. 7–8
- The central governance deficit is the absence of red lines for uncontrollable AI, while transparency and accountability demands address gaps that existing norms may otherwise cover.The review distinguishes these structural challenges from issues such as bias and product safety, which it says can be addressed through existing human-rights norms and product-liability law.Source: §2 · p. 8
- Although Maas frames AI governance as a collective-action problem and potential global public good, the review questions whether governance can be handled by an undifferentiated “we.”The review distinguishes the collective case for cooperation from explaining why states and technology companies with power in this domain would create governance.Source: §2 · pp. 8, 11
- Military AI governance has advanced through UN forums since 2014, but definitional disputes have stalled progress and willingness to commit varies inversely with deployment capacity.The review notes that the CCW Group of Government Experts has met annually since 2017, with progress remaining halting.Source: §2 · p. 9
- Governance efforts shifted from conventional AI discussions and state regulation to advanced-model initiatives, while disputes such as the Pentagon-Anthropic clash blur community boundaries.The EU AI Act was adopted in 2024; the UN's Global Digital Compact established policy dialogues and a scientific panel.Source: §2 · pp. 9–10
- Much global normative movement remains non-binding, though the review notes that broad participation with shallow obligations has also characterized the development of human-rights norms.The review treats non-binding commitments as not necessarily undesirable.Source: §2 · p. 10
- International discussion is underinclusive, with the Global South receiving one substantive paragraph and China only a few pages, despite equity and effectiveness arguments for wider participation.The review points to the global reach of AI-related collective-action problems and to India's emphasis on impact and benefit distribution.Source: §2 · p. 11
3 The Trillion Dollar Question
The review argues that AI governance is already an emerging architecture, but its design cannot be separated from the distribution of public and private power. States retain important levers, while many consequential frontier-AI decisions sit inside private firms.
3 The Trillion Dollar Question
- The reviewer concedes that this review may share omissions and points readers to further work on digital politics and AI governance in Africa and the Global Majority.The cited works include African AI governance and a report titled AI from the Global Majority.Source: §3 · p. 11
- The Global Digital Compact contrasts states’ commitment to align relevant digital legislation with international law with a call for companies to respect human rights.The review identifies this as a challenge in implementing state-adopted norms in the private sector.Source: §3 · pp. 11–12
- Frontier AI’s consequential decisions and key chokepoints are concentrated in private firms, while states can regulate markets, control strategic inputs, and impose costs [47].The review names compute, cloud, model development, and deployment as concentrated chokepoints.Source: §3 · pp. 11–12
- Washington’s advanced-chip controls remained, while the succeeding administration declined to enforce and moved to rescind the AI Diffusion Rule, illustrating the contingency of state power.The passage also says the administration retained and adapted controls on advanced chips.Source: §3 · p. 12
- Maas offers no single roadmap; his strategies section instead proposes a toolbox for evaluating regimes’ efficacy, resilience, and coherence.He avoids prescribing substantive values or goals for AI regimes because they may change over time.Source: §3 · p. 13