Matthijs Michiel Maas

Global catastrophic risks ("GCRs") are risks capable of producing significant and severe harms affecting the human rights, interests, or well-being of tens or hundreds of millions of people across the globe. This chapter examines global catastrophic risks as a challenge for international governance. It evaluates risk definitions, traces the historical occurrences of global catastrophes, and explains why risks of shared concern can nevertheless prove difficult to mitigate because of political, psychological, and institutional barriers. It then surveys the range of scientific, environmental, or technological phenomena that have been proposed as sources of global catastrophic risk, before developing a framework for identifying which risks are sufficiently direct, globally scalable, and severe in their impacts as to constitute global catastrophic risks, and which are regulated under international law. The chapter concludes that the regimes on biological weapons, nuclear weapons, pandemics, ozone layer depletion, and extreme climate change can provide risk-specific legal guidance.

International law has long been driven by the imperative to prevent mass harm to humanity. In a new era of global catastrophic risks, including from novel sources such as advanced artificial intelligence, it can and must continue to serve as a cornerstone for mitigating these epochal threats to human rights, safety, and welfare. This Introduction establishes the book's core argument: that States are already legally obligated under international law to mitigate all global catastrophic risks. These obligations can be drawn from treaties and customary international law, from risk-specific regimes that target particular global hazards as well as cross-cutting regimes protecting humans or their environment. These obligations can be interpreted and applied to the mitigation of catastrophic risks from advanced AI through systemic integration, drawing on general principles of law derived from the international legal system. This introduction clarifies the doctrinal and political foundations of this argument, articulates its limits, and compares it to other strategies for mitigating global catastrophic risks.

How will AI disrupt the substance, processes, or political scaffolding of law, regulation, and governance? This chapter draws on the concept of governance disruption, from the field of technology law. It explores how new AI capabilities can produce intended and unintended changes in international law and in the broader global governance architecture. After defining ‘governance disruption’, and illustrating the historical connections between technological innovation and legal change, it makes the case that we have strong reasons to expect international law to be especially susceptible to disruption by AI. The chapter then introduces a framework for mapping governance disruption, before reviewing, in turn, three types of such disruption. In cases of (1) governance development, the sociotechnical changes produced by AI systems result in substantive legal uncertainties and ambiguities, which need resolution through legal adaptation to ensure that laws and institutions remain fit for purpose. In cases of (2) governance displacement, AI technologies may be used to support key processes in law-making and law enforcement, or even to substitute for them, with both beneficial and adverse effects. In cases of (3) governance destruction, AI uses could erode the effectiveness and coherence of the global legal order, either because existing international regimes prove conceptually or politically incapable of carrying through pressing legal developments or because AI systems directly or indirectly erode the foundational political scaffolding of those regimes. Consequently, the chapter argues that AI governance efforts must come to terms with the dynamics of governance disruption, to ensure they are resilient to future normative, substantive, and political challenges they are to face arising from the use of AI.

The global AI governance architecture is in a state of rapid change. How can we understand the origins, shape, and future evolution of this landscape? How can we anticipate the consequences of different institutional trajectories, and develop effective governance strategies in response, to manage these? This chapter aims to show how the three conceptual frameworks worked out in this book—sociotechnical change, governance disruption, and regime complexity—can help answer these questions. These concepts help us reframe key questions around global AI governance, offering new perspectives on the choices we face in determining its future directions. Specifically, they offer insight into five questions. In terms of an existing, emerging, or proposed AI regime’s origins, we can study an institution’s intended purpose and political foundations, to see how these may affect that regime’s optimal design—and its likely default design. In terms of topology, we can consider where a specific institution fits in the broader normative and institutional structure of the broader regime complex. In terms of its evolution, we can examine likely future trajectories of that complex, as driven by external governance trends and AI-induced governance disruptions. In terms of consequences, we can consider the political, institutional and normative challenges, compromises and trade-offs that the AI governance architecture may face, conditional on its evolutionary trajectory tending towards fragmentation or centralization. Finally, in terms of strategies, these frameworks can help us consider appropriate shifts in our general regulatory approach, instrument choice, and in instrument design, which can help ensure that the AI regime complex will achieve adequate efficacy, resilience, and coherence in the face of global technological, legal and political change.

How can we govern a changing technology, in a changing world, while using global governance instruments that may themselves be left changed by that very technology? This introduction both grounds and summarizes the book and its method, content, argument, and structure. It sets the stage for the global AI governance challenge, discussing recent progress in AI technology, the resulting spectrum of global challenges, recent and ongoing developments and initiatives in global AI governance, and the political hurdles that likely lie ahead. It reviews the development of the academic field of advanced AI governance, and argues that, to act in a timely way, scholars and policymakers need to reckon with three principal facets of structural change in AI governance: change in AI’s sociotechnical problems; change in the tools of governance; and change in its political and institutional environment. It therefore introduces and briefly discusses three conceptual frameworks, drawn from the fields of technology law and global governance scholarship, that can facilitate this reckoning to the global governance of AI. It presents the book’s central thesis: that we can and must choose to adapt our global governance architectures for AI in a manner that is more robust to ongoing and future changes in AI technology, in law, and in international politics.

How does AI technology drive social change? When, where, and why are such changes relevant for law? How are they best regulated? This chapter argues that AI governance research and policymaking can gain from a focus on sociotechnical change. This framework can clarify precisely what it is we seek to regulate, why regulation is needed, when to pursue regulation, and how to tailor such regulation to different governance targets across the AI development and deployment stack. To do so, the chapter first reviews four prevailing regulatory approaches to technology, which respectively focus on: the new technology as an exceptional class; its varied applications; its impacts on particular values; and its challenges in existing legal domains. As an alternative, it argues for a regulatory approach focused on sociotechnical change. Drawing on the field of technology law, the chapter precisely defines ‘technology’ and ‘sociotechnical change’, in order to discuss the mutually constitutive relationship between technological change and societal change—and the scope of sociotechnical choice they afford—through historical examples. It then tracks how orienting AI governance around sociotechnical change should open a path to more effective AI regulation, by aiding our understanding of: (1) why to regulate; (2) when to regulate, given prevailing uncertainty over AI’s impacts; and (3) how to regulate different AI governance targets, taking into account both relevant material features of the AI models and their infrastructure, as well as the sociotechnical problem logics their uses engage. Finally, the chapter reviews the analytical uses and limits of the sociotechnical perspective on AI governance. While this approach is not a panacea, governance efforts for AI must reckon with sociotechnical change to ensure their efficacy.

As AI systems have become increasingly capable and widely used, calls have grown for globally coordinated governance of this powerful technology, to check its potentially runaway impacts. Yet what are the precise issues in play? Why would they need global governance? Can any regulations be globally enforced? In response to such questions, this chapter explores the necessity, urgency, and enforceability of global AI governance. It reviews existing taxonomies for AI’s potential impacts and policy challenges, to suggest that while the spectrum of challenges posed by AI is broad, it is in fact possible to organize global AI governance debates along both technical and political axes. As such, we can distinguish three global governance focus areas for mitigating the challenges of conventional (narrow) AI, military AI, and advanced AI, respectively. Many of these AI issue clusters are global public good problems, such that their effective resolution will gain from—or may require—international cooperation. Reviewing models for the global regulation of digital technologies, the chapter argues that, in spite of political challenges and barriers to establishing AI regimes, once created such arrangements can avail themselves of many regulatory levers in order to effectively govern crucial digital objects and components across the development chain of AI systems. Consequently, global agreements on AI would be broadly enforceable, provided that such regimes are established as soon as possible. Scaling international law for the challenges of AI is both necessary and urgent, if the world is to meet AI’s diverse challenges in a timely manner.

What is the state of global AI governance? Where is it set to go next? This chapter charts the history of the global AI governance architecture (2009–2025), and its potential future pathways. It maps three main tracks in global initiatives and norm development in AI governance. These tracks have focused on governing military AI systems (2009–), conventional narrow AI systems (2015–), and increasingly advanced general-purpose AI systems (2022–), respectively. The chapter evaluates both the strengths of these institutional achievements to date, and outstanding weaknesses in this emerging global governance architecture. While recent years have seen remarkable progress, this governance architecture remains fragmented. A critical risk facing us now is that these governance initiatives may not consolidate in time to adequately address all three issue clusters, or even any one. In response, the chapter reviews a range of frequently proposed pathways forward, including: the extension of non-binding soft law; unilateral state actions or transboundary regulations; the application of existing norms of international law; the reform of existing governance institutions; the negotiation of new international agreements; and, especially, the creation of potential new international organizations for AI, along one of seven common models. While many of these tools are promising, their absolute or comparative efficacy remains uncertain. New insights or strategies may be needed to help improve the global AI governance architecture in ways that secure a human future.