2.1. What is ‘systems thinking’ and why does it matter for global risk?

‘Systems thinking’ recognises that a system is more than the sum of its parts (Meadows, Reference Meadows2008): its behaviour is emergent and cannot be predicted purely from its constituent components.Footnote ⁴ More specifically, CAS theory describes systems in which patterns and adaptive behaviours at higher levels emerge from localised interactions and selection at lower levels (Holland, Reference Holland1995; Levin, Reference Levin1998; Levin et al., Reference Levin, Xepapadeas, Crépin, Norberg, De Zeeuw, Folke, Hughes, Arrow, Barrett, Daily, Ehrlich, Kautsky, Mäler, Polasky, Troell, Vincent and Walker2013; Miller & Page, Reference Miller and Page2009). Examples of CASs include ecosystems (Levin, Reference Levin1998), social systems (Miller & Page, Reference Miller and Page2009), economies (Arthur, Reference Arthur2021; Levin et al., Reference Levin, Xepapadeas, Crépin, Norberg, De Zeeuw, Folke, Hughes, Arrow, Barrett, Daily, Ehrlich, Kautsky, Mäler, Polasky, Troell, Vincent and Walker2013), food systems (Chapman et al., Reference Chapman, Klassen, Kreitzman, Semmelink, Sharp, Singh and Chan2017), infrastructure systems (Oughton et al., Reference Oughton, Usher, Tyler and Hall2018), social–ecological systems more broadly (Levin et al., Reference Levin, Xepapadeas, Crépin, Norberg, De Zeeuw, Folke, Hughes, Arrow, Barrett, Daily, Ehrlich, Kautsky, Mäler, Polasky, Troell, Vincent and Walker2013; Preiser et al., Reference Preiser, Biggs, De Vos and Folke2018), and the Earth's entire biosphere (Folke et al., Reference Folke, Polasky, Rockström, Galaz, Westley, Lamont, Scheffer, Österblom, Carpenter, Chapin, Seto, Weber, Crona, Daily, Dasgupta, Gaffney, Gordon, Hoff, Levin and Walker2021; Levin, Reference Levin1998). CASs can behave in highly non-linear ways: large changes can have small effects, small changes can have large effects, and disruptions in one part of the system can cascade into other parts. Broadly, systems thinking allows us to make sense of such behaviours; this matters for global risk and GCR because many of the systems that humans exist within and depend on (Avin et al., Reference Avin, Wintle, Weitzdörfer, Ó hÉigeartaigh, Sutherland and Rees2018) are CASs (see also Section 2.2).

Our usage of ‘complex’ here is not synonymous with ‘complicated’ (Kreienkamp & Pegram, Reference Kreienkamp and Pegram2020; Miller & Page, Reference Miller and Page2009; Snowden & Boone, Reference Snowden and Boone2007; UNDRR, 2019). A system can be complicated but not complex: this is the case if it is intricate and difficult to understand, but still largely comprehensible based on its component parts and predictable relationships between them (i.e. emergent system properties do not play an important role). A useful example of such a system is a jet engine (Kreienkamp & Pegram, Reference Kreienkamp and Pegram2020).

Much emergent behaviour in systems can be described using the language of feedbacks. Feedbacks occur when the state of a system affects (feeds back on) the state of that same system, either directly or indirectly. Mathematically, positive feedbacks (+) are destabilising (a given change causes more of that change to occur), while negative feedbacks (−) are stabilising (the effects of a given change can be damped out). This is summarised in Figure 1, using ‘causal loop’ and ‘stability landscape’ diagrams. Both kinds of feedbacks can have desirable and undesirable effects: stabilising feedbacks can preserve desirable system states but also trap systems in undesirable states, while destabilising feedbacks can drive runaway evolution to desirable or undesirable system states. Most systems of interest contain multiple feedbacks, which can change in strength depending on external or internal factors and can lead to sudden shifts in system behaviour (e.g. when an equilibrium state changes stability).

Figure 1.

Summary of destabilising (mathematically positive) and stabilising (mathematically negative) feedbacks. In causal loop diagrams (top row), an arrow with a + symbol means that a change in the first variable causes the second variable to change in the same direction (e.g. an increase in A causes an increase in B), and an arrow with a − symbol means that a change in the first variable causes the second variable to change in the opposite direction (e.g. an increase in A causes a decrease in B). In stability landscape diagrams (bottom row), the state of the system is conceived of as a ball rolling on a landscape (collapsing the high-dimensional state spaces of the real world onto a single dimension). When destabilising feedbacks dominate, we see runaway change (rolling down the hill); when stabilising feedbacks dominate, the system remains within a stable equilibrium (the valley, or ‘basin of attraction’).

Another valuable concept for understanding CAS dynamics is that of resilience. While early work characterised resilience simply as the size of the system's current basin of attraction (Holling, Reference Holling1973; Scheffer et al., Reference Scheffer, Carpenter, Foley, Folke and Walker2001), Walker et al. (Reference Walker, Holling, Carpenter and Kinzig2004) define it more qualitatively as ‘the capacity of a system to absorb disturbance and reorganise while undergoing change so as to still retain essentially the same function, structure, identity, and feedbacks’. Resilience is deeply related to adaptability (the capacity of a system to adjust its responses based on changing conditions) and transformability (the capacity to transform the stability landscape to become a new kind of system); critically, maintaining system identity at large scales under changing conditions may require radical system transformations at smaller scales (Folke et al., Reference Folke, Carpenter, Walker, Scheffer, Chapin and Rockström2010).

This understanding of resilience as a dynamic property involving transformation and reorganisation naturally leads us to consider how such changes unfold over time. One theoretical framework that explicitly addresses these temporal dynamics is adaptive cycle theory (Gunderson & Holling, Reference Gunderson and Holling2002). Building on examples from ecology, this highlights that many systems undergo a cycle of: exploitation, in which there is rapid expansion of agents or strategies into new areas; conservation, in which resources slowly accumulate and the system becomes more interconnected and rigid; release, in which the rigid interconnected system fails and collapses; and reorganisation, in which the system is restructured and new kinds of agents or strategies arise. In practice, systems may exhibit a ‘panarchy’ of nested adaptive cycles on a range of scales (Allen et al., Reference Allen, Angeler, Garmestani, Gunderson and Holling2014; Gunderson et al., Reference Gunderson, Allen and Garmestani2022; Gunderson & Holling, Reference Gunderson and Holling2002).

2.2. The global system

To see how systems thinking can inform our understanding of global risk, it is worth developing a notion of ‘the global system’. By the ‘global system’, we mean the globally interconnected system of human economic, social, political, and cultural relations, including humans themselves, material flows, and the extraction of materials from the broader Earth system. Our conceptualisation draws from both world-systems theory in the social sciences (Chase-Dunn & Grimes, Reference Chase-Dunn and Grimes1995) and from Earth system science, where the global system has been labelled the ‘Anthroposphere’ (Steffen et al., Reference Steffen, Richardson, Rockström, Schellnhuber, Dube, Dutreuil, Lenton and Lubchenco2020). While the global system is a subsystem of the Earth system, the boundaries are fuzzy: the two systems have substantially shaped one another (Ellis, Reference Ellis2015; Frankopan, Reference Frankopan2023; Nyström et al., Reference Nyström, Jouffray, Norström, Crona, Søgaard Jørgensen, Carpenter, Bodin, Galaz and Folke2019; Williams et al., Reference Williams, Zalasiewicz, Haff, Schwägerl, Barnosky and Ellis2015). The global system is a CAS, and so are many of its subsystems.

Why is this concept useful? For our purposes, there are two main reasons. First, by identifying a single global system within which most humans are embedded and upon which most humans rely, we can recognise that this system mediates almost all GCR.Footnote ⁵ This is an important insight. Early work on GCR often treated risk as essentially synonymous with hazard: the risk of an asteroid impact or the risk of dangerous climate change. But this neglects the important role played by vulnerability: how will the global system respond when stressed by a hazard? Foregrounding this can drastically change our understanding of GCR.

Second, identifying the global system as a CAS informs us that it will display emergent phenomena, such as non-linear behaviour, path dependence, feedbacks, and cascading failures. We need to understand these phenomena to understand and mitigate global risk.

We will occasionally find it convenient to refer to the global system as a single entity, which, for example, ‘creates’, ‘generates’, or ‘amplifies’ various drivers of risk. This is meant neither to anthropomorphise the global system nor to suggest that all parts of the global system are equally responsible for the phenomenon being discussed. It is also not intended to remove culpability or agency from individual actors, such as people, states, or corporations. As one illustrative example, in the case of climate change, it is simultaneously true that a small number of companies conduct the extraction of fossil fuel leading to the vast majority of carbon dioxide emissions (Carbon Majors, 2024; Heede, Reference Heede2014), that these companies exist within a global system within which it is highly profitable to extract and sell fossil fuels, and that the demand for the energy produced ultimately derives from the constrained choices of billions of individual human beings.

2.3. Emergent global risk: a wide array of concepts

In this section, we review some of the many concepts used to describe how risk emerges from the complex global system. A natural starting point is with ‘systemic risk’. Early definitions of the term (Kaufman & Scott, Reference Kaufman and Scott2003; OECD, 2003) emphasised the risk of failure in an entire system, as opposed to failures in some of its components. Subsequent work has emphasised the risk of smaller disruptions being amplified, for example, due to non-linearity, interconnectedness, and cascading failure (Centeno et al., Reference Centeno, Nag, Patterson, Shaver and Windawi2015; May et al., Reference May, Levin and Sugihara2008; Renn et al., Reference Renn, Lucas, Haas and Jaeger2017, Reference Renn, Laubichler, Lucas, Kröger, Schanze, Scholz and Schweizer2022; Sillmann et al., Reference Sillmann, Christensen, Hochrainer-Stigler, Huang-Lachmann, Juhola, Kornhuber, Mahecha, Mechler, Reichstein, Ruane, Schweizer and Williams2022). The concepts of ‘global systemic risk’ (Centeno et al., Reference Centeno, Nag, Patterson, Shaver and Windawi2015; Renn et al., Reference Renn, Lucas, Haas and Jaeger2017) and ‘hyper-risks’ (Helbing, Reference Helbing2013) apply these ideas specifically to the global system. Even more specifically, the concept of ‘femtorisk’ focuses on the systemic risk due to ‘the actions and interactions of actors existing beneath the level of formal institutions, often operating outside effective governance structures’ (Frank et al., Reference Frank, Collins, Levin, Lo, Ramo, Dieckmann, Kremenyuk, Kryazhimskiy, Linnerooth-Bayer, Ramalingam, Roy, Saari, Thurner and von Winterfeldt2014).

There is a wider set of related and often overlapping concepts from risk analysis, including compound risk, interacting risk, interconnected risk, and cascading risk (Pescaroli & Alexander, Reference Pescaroli and Alexander2018). Although discussion of the nuances is beyond the scope of this section, we do note that the concept of compound risk – the risk of disasters involving multiple simultaneously occurring hazards (Pescaroli & Alexander, Reference Pescaroli and Alexander2018; Zscheischler et al., Reference Zscheischler, Westra, Van Den Hurk, Seneviratne, Ward, Pitman, AghaKouchak, Bresch, Leonard, Wahl and Zhang2018, Reference Zscheischler, Martius, Westra, Bevacqua, Raymond, Horton, van den Hurk, AghaKouchak, Jézéquel, Mahecha, Maraun, Ramos, Ridder, Thiery and Vignotto2020) – has recently been applied more broadly to global systems in light of the COVID-19 pandemic (Kruczkiewicz et al., Reference Kruczkiewicz, Klopp, Fisher, Mason, McClain, Sheekh, Moss, Parks and Braneon2021). If the co-occurring hazards cause more damage together than had they occurred separately, this is precisely an instance of non-linear amplification – a key characteristic of CASs.

The global system also interacts with the global environment to produce risk. Human actions are by far the dominant driver of global environmental change (Ellis, Reference Ellis2015) and could trigger a wide range of ecological and environmental tipping points (Armstrong McKay et al., Reference Armstrong McKay, Staal, Abrams, Winkelmann, Sakschewski, Loriani, Fetzer, Cornell, Rockström and Lenton2022; Barnosky et al., Reference Barnosky, Hadly, Bascompte, Berlow, Brown, Fortelius, Getz, Harte, Hastings, Marquet, Martinez, Mooers, Roopnarine, Vermeij, Williams, Gillespie, Kitzes, Marshall, Matzke and Smith2012; Lenton et al., Reference Lenton, Held, Kriegler, Hall, Lucht, Rahmstorf and Schellnhuber2008; Richardson et al., Reference Richardson, Steffen, Lucht, Bendtsen, Cornell, Donges, Drüke, Fetzer, Bala, von Bloh, Feulner, Fiedler, Gerten, Gleeson, Hofmann, Huiskamp, Kummu, Mohan, Nogués-Bravo and Rockström2023). One useful conceptual framework here is that of ‘Anthropocene risks’ (Keys et al., Reference Keys, Galaz, Dyer, Matthews, Folke, Nyström and Cornell2019), which are described as originating from anthropogenic Earth system change, emerging due to the evolution of globally intertwined social–ecological systems, and involving complex cross-scale interactions.

A separate and emerging body of work applies the concept of tipping points – where a small change can have a large, abrupt, self-perpetuating, and hard-to-reverse impact – to the context of negative social change (Juhola et al., Reference Juhola, Filatova, Hochrainer-Stigler, Mechler, Scheffran and Schweizer2022; Spaiser et al., Reference Spaiser, Juhola, Constantino, Guo, Watson, Sillmann, Craparo, Basel, Bruun, Krishnamurthy, Scheffran, Pinho, Okpara, Donges, Bhowmik, Yasseri, Safra de Campos, Cumming, Chenet and Spears2024).Footnote ⁶ Such ‘negative social tipping points’ are another way to understand the non-linear creation of risk in the global system. While the focus thus far has been on stresses from anthropogenic climate change, this does not need to remain the case: tipping points in the global system could be triggered by a wide range of factors.

The subject of negative social tipping points leads us straightforwardly to collapse. Societal collapse (Brozović, Reference Brozović2023; Centeno et al., Reference Centeno, Callahan, Larcey and Patterson2023; Tainter, Reference Tainter1988) likely involves interacting non-linear processes (e.g. tipping) within the system itself (Centeno et al., Reference Centeno, Callahan, Larcey, Patterson, Izdebski, Haldon and Filipkowski2022; Cumming & Peterson, Reference Cumming and Peterson2017; Homer-Dixon, Reference Homer-Dixon2006; Juhola et al., Reference Juhola, Filatova, Hochrainer-Stigler, Mechler, Scheffran and Schweizer2022; Lenton, Reference Lenton, Centeno, Callahan, Larcey and Patterson2023). There is a natural intersection between the study of collapse and the study of GCR and worst-case outcomes (Belfield, Reference Belfield, Centeno, Callahan, Larcey and Patterson2023). Global collapse is an important form of global catastrophe. A collapse is usually preceded by a crisis (Butzer, Reference Butzer2012).

The ‘synchronous failure’ framework (Homer-Dixon et al., Reference Homer-Dixon, Walker, Biggs, Crépin, Folke, Lambin, Peterson, Rockström, Scheffer, Steffen and Troell2015) offers a more specific causal description of how crises emerge in the modern global system. The authors first identify three important global trends: the scale of human activity, increased connectivity, and reduced diversity. They then argue that these trends favour three specific ‘process archetypes’ – ‘long fuse big bang’, ‘simultaneous stresses’, and ‘ramifying cascade’ – which interact within and across systems to produce global crises. These archetypes are relevant not just for the modern global system, but have also been applied historically, for example to the Late Bronze Age Collapse (Kemp & Cline, Reference Kemp, Cline, Izdebski, Haldon and Filipkowski2022).

Finally, the concept of polycrisis captures the general idea that the world's crises are coinciding and converging into a whole worse than the sum of its parts (Lawrence et al., Reference Lawrence, Homer-Dixon, Janzwood, Rockstöm, Renn and Donges2024; Morin & Kern, Reference Morin and Kern1999; Tooze, Reference Tooze2021). Lawrence et al. (Reference Lawrence, Homer-Dixon, Janzwood, Rockstöm, Renn and Donges2024) offer a more specific definition: global polycrisis is the ‘causal entanglement of global crises in ways that significantly degrade humanity's prospects’. They describe this entanglement by distinguishing between stresses, triggers, and crises (Section 5.2), and considering how these interact within and across systems. In this article, we typically use ‘polycrisis’ to refer to this specific conceptual framework.

3.1. Hazards

As outlined earlier, we define hazards as the events and processes that can serve as the proximal causes of undesirable outcomes. While early discussions of GCR often focused on exogenous hazards like asteroid impacts, many of the most concerning hazards facing humanity emerge from within the global system. For instance, anthropogenic climate change occurs due to humanity's consumption of fossil fuels (IPCC, Reference Masson-Delmotte, Zhai, Pirani, Connors, Péan, Berger, Caud, Chen, Goldfarb, Gomis, Huang, Leitzell, Lonnoy, Matthews, Maycock, Waterfield, Yelekçi, Yu and Zhou2021). Technological as well as geopolitical developments led to the creation of nuclear weapons, the maintenance of nuclear arsenals, and thus the risk of global nuclear war (Jacobsen, Reference Jacobsen2024; Waltz & Sagan, Reference Waltz and Sagan1995). Environmental encroachment promotes the emergence of new pandemic hazards (Jones et al., Reference Jones, Grace, Kock, Alonso, Rushton, Said, McKeever, Mutua, Young, McDermott and Pfeiffer2013; Singh, Reference Singh2021), and advances in biotechnology may do the same in the future (Millett & Snyder-Beattie, Reference Millett and Snyder-Beattie2017; Musunuri et al., Reference Musunuri, Sandbrink, Monrad, Palmer and Koblentz2021). Despite many concerns about the societal impacts of AI, the development of more and more powerful AI models is essentially racing full steam ahead, in large part due to economic and geopolitical dynamics (Brandt et al., Reference Brandt, Kreps, Meserole, Singh and Sisson2022; Lee, Reference Lee2018).

Why are these hazards being created? One possible answer focuses on the actions of agents. For example, Kemp (Reference Kemp2021) has argued that a small number of ‘agents of doom’ pursue power and profit at the expense of creating these hazards for the rest of humanity. Seventy-eight corporate and state fossil fuel-producing entities are responsible for more than 70% of total cumulative carbon dioxide emissions (Carbon Majors, 2024), only nine states possess nuclear weapons (Herre et al., Reference Herre, Rosado, Roser and Hasell2024), and only a handful of companies are currently leading the ‘AI arms race’.Footnote ⁹ Understanding how responsibility for hazard creation may be concentrated among a small number of actors is important not only for its own sake but also for its instrumental value: it can provide important insights about ways to lower GCR (Jones, Reference Jones, Beard, Rees, Richards and Rojas2023).

Nevertheless, part of the problem is more pernicious: these actors exist within systems in which it makes sense – or appears to make sense – for them to create risk. Governments justify their maintenance of nuclear arsenals based on principles of strategic rationality (Amadae, Reference Amadae2015). Companies extract and sell fossil fuels because they can make a large profit doing so, and individuals are constrained and incentivised to consume them. One approach for describing this behaviour is using game theory: in particular, the ‘tragedy of the commons’ highlights that it can seem individually rational to help deplete certain resources, including our collective security from global risk (Barrett, Reference Barrett2016; Posner, Reference Posner2004). However, such rational-actor narratives have key limitations (Dietz et al., Reference Dietz, Ostrom and Stern2003; Ostrom, Reference Ostrom1990) and can also serve as part of the problem, by themselves encouraging hazard-creating actions (Amadae, Reference Amadae2015).

Going a step further, the incentives for actors to create global hazards exist in part because national and international institutions have not taken effective regulatory action to prevent these incentives from existing. While this is in part because these institutions themselves have little incentive to understand and prepare for large-scale unprecedented events (Posner, Reference Posner2004; Wiener, Reference Wiener2016), it is also due to a number of other issues: some are discussed below, and others in Section 4. In any case, while actors remain free to create global hazards for power and profit, some will choose to do so.

Hazard-creating systems can also entrench themselves in ways that make change difficult. For example, in the case of climate change, fossil fuel companies sowed a sophisticated, decades-long disinformation campaign to mislead the public (Oreskes & Conway, Reference Oreskes and Conway2011; Supran et al., Reference Supran, Rahmstorf and Oreskes2023). They are also supported by a range of governmental subsidies and tax incentives, which actively make decarbonisation less economically feasible (Seto et al., Reference Seto, Davis, Mitchell, Stokes, Unruh and Ürge-Vorsatz2016). At a systems level, there are effectively stabilising feedbacks in play: the fossil fuel–industrial complex is deeply resistant to change. Understanding these kinds of mechanisms is vital for hazard reduction.

Ultimately, these patterns of hazard creation reflect a fundamental fact about CASs. Because selection (e.g. biological, cultural, or economic) occurs at lower levels (Levin, Reference Levin1998; Levin et al., Reference Levin, Xepapadeas, Crépin, Norberg, De Zeeuw, Folke, Hughes, Arrow, Barrett, Daily, Ehrlich, Kautsky, Mäler, Polasky, Troell, Vincent and Walker2013), the behaviours which are selected for are not necessarily beneficial to the system as a whole. This has also been described using the concept of ‘evolutionary traps’ (Søgaard Jørgensen et al., Reference Søgaard Jørgensen, Jansen, Avila Ortega, Wang-Erlandsson, Donges, Österblom, Olsson, Nyström, Lade, Hahn, Folke, Peterson and Crépin2024). In this case, some of the selected behaviours create global hazards.

3.2. Amplification

The global system can also amplify hazards. Due to the process of amplification, a realised hazard or set of hazards may cause a catastrophe much larger than the direct impact itself. The March 2021 blocking of the Suez Canal by the container ship Ever Given, with consequences of billions of US dollars in lost trade income (Russon, Reference Russon2021), was a clear demonstration of how small hazards can be amplified within the global system. More generally, the possibility of non-linear (i.e. disproportionate) amplification of risk by the global system or its many subsystems has been the focus of much of the work reviewed in Section 2.3 (Centeno et al., Reference Centeno, Nag, Patterson, Shaver and Windawi2015; Frank et al., Reference Frank, Collins, Levin, Lo, Ramo, Dieckmann, Kremenyuk, Kryazhimskiy, Linnerooth-Bayer, Ramalingam, Roy, Saari, Thurner and von Winterfeldt2014; Helbing, Reference Helbing2013; Homer-Dixon et al., Reference Homer-Dixon, Walker, Biggs, Crépin, Folke, Lambin, Peterson, Rockström, Scheffer, Steffen and Troell2015; Lawrence et al., Reference Lawrence, Homer-Dixon, Janzwood, Rockstöm, Renn and Donges2024; Renn et al., Reference Renn, Lucas, Haas and Jaeger2017, Reference Renn, Laubichler, Lucas, Kröger, Schanze, Scholz and Schweizer2022; Sillmann et al., Reference Sillmann, Christensen, Hochrainer-Stigler, Huang-Lachmann, Juhola, Kornhuber, Mahecha, Mechler, Reichstein, Ruane, Schweizer and Williams2022; Spaiser et al., Reference Spaiser, Juhola, Constantino, Guo, Watson, Sillmann, Craparo, Basel, Bruun, Krishnamurthy, Scheffran, Pinho, Okpara, Donges, Bhowmik, Yasseri, Safra de Campos, Cumming, Chenet and Spears2024).

For a simple conceptual understanding of amplification in the context of GCR, we again consider the simple metaphor of a ball in a stability landscape (Figure 3). We conceive of the global system as existing in a state of dynamic quasi-equilibrium: it is constantly evolving, yet its basic function and structure persist, and minor shocks can be recovered from. Yet other possible system states exist, some of which would constitute global catastrophic outcomes. Hazards (Section 3.1) can push the system out of its current basin of attraction towards one of these (normatively) worse states, whereupon the effect of the hazard is amplified by feedbacks internal to the system. Vulnerability enters this picture in multiple ways: how shallow is the original basin of attraction (how weak are the initial stabilising feedbacks), how strong are the feedbacks driving the system to the new outcome, and how severe (i.e. normatively bad) is the new outcome? The concept of resilience also relates to the depth of the original basin of attraction, but in the opposite direction: here, a decrease in resilience is an increase in vulnerability.

Figure 3.

Amplification in the context of global catastrophic risk. Hazards (modulated by exposure) threaten the system's persistence in its current basin of attraction and can set in motion runaway evolution towards global catastrophic outcomes (amplification). The broad notion of vulnerability relates to multiple things here: how deep are the two basins of attraction, and how bad is the global catastrophic outcome? We emphasise that this picture is vastly oversimplified (see Section 3.2), but it captures important elements of the problem.

This picture is of course an oversimplification. There are many possible kinds of global catastrophic outcomes, many possible pathways to them, and the degree of amplification will vary based on the specific case. Figure 3 depicts amplification as occurring via a tipping point,Footnote ¹⁰ but this does not need to be the case: amplification can also occur without tipping points. Amplification can also occur when multiple co-occurring hazards cause more damage together than they would have separately, or when one hazard triggers another hazard. The key point is that emergent behaviour in CASs can lead to an amplification of hazards. Figure 3 highlights this and also allows us to make important conceptual connections.

Amplification would play an important role in a wide range of global catastrophe scenarios. The impacts of natural hazards like asteroid impacts and volcanic eruptions would be amplified through their effects on global critical infrastructure (Baum, Reference Baum2023; Mani et al., Reference Mani, Erwin, Johnson, Beard, Rees, Richards and Rojas2023; Moersdorf et al., Reference Moersdorf, Rivers, Denkenberger, Breuer and Jehn2023); indeed, the clustering of such infrastructure near centres of volcanic activity vastly amplifies the risk even from lower-magnitude volcanic eruptions (Mani et al., Reference Mani, Tzachor and Cole2021). As the COVID-19 pandemic has illustrated, the impact of novel infectious diseases can be vastly amplified by global transit networks (Baker et al., Reference Baker, Mahmud, Miller, Rajeev, Rasambainarivo, Rice, Takahashi, Tatem, Wagner, Wang, Wesolowski and Metcalf2022) as well as by follow-on economic and social disruption. The worst outcomes from climate change will likely not arise directly due to increased temperatures, but rather indirectly through phenomena like conflict, famine, and mass displacement (Beard et al., Reference Beard, Holt, Tzachor, Kemp, Avin, Torres and Belfield2021; Kemp et al., Reference Kemp, Xu, Depledge, Ebi, Gibbins, Kohler, Rockström, Scheffer, Schellnhuber, Steffen and Lenton2022; Richards et al., Reference Richards, Lupton and Allwood2021). The effects of nuclear war could be amplified by subsequent global cooling, leading to large-scale global starvation (Xia et al., Reference Xia, Robock, Scherrer, Harrison, Bodirsky, Weindl, Jägermeyr, Bardeen, Toon and Heneghan2022). Developments in AI capabilities could be amplified towards catastrophic outcomes well before the emergence of artificial general intelligence, for example, via societal and economic destabilisation (Kasirzadeh, Reference Kasirzadeh2025; Kulveit et al., Reference Kulveit, Douglas, Ammann, Turan, Krueger and Duvenaud2025) and interactions with biological and nuclear risk (Bengio et al., Reference Bengio, Mindermann, Privitera, Besiroglu, Bommasani, Casper, Choi, Fox, Garfinkel, Goldfarb, Heidari, Ho, Kapoor, Khalatbari, Longpre, Manning, Mavroudis, Mazeika, Michael and Zeng2025; EBRC, 2023; Hendrycks et al., Reference Hendrycks, Mazeika and Woodside2023; Maas et al., Reference Maas, Lucero-Matteucci, Cooke, Beard, Rees, Richards and Rojas2023)

The work on emergent global risk (Section 2.3) helps us understand further details of how the global system can amplify hazards. Amplification on networks can be understood in terms of contagion and cascading failure (Helbing, Reference Helbing2013; Krönke et al., Reference Krönke, Wunderling, Winkelmann, Staal, Stumpf, Tuinenburg and Donges2020; Newman, Reference Newman2018) and has played a key role in conceptualisations of global systemic risk (Centeno et al., Reference Centeno, Nag, Patterson, Shaver and Windawi2015). Many of the amplifying effects alluded to in the above paragraph result from non-linear social dynamics (Miller & Page, Reference Miller and Page2009; Schelling, Reference Schelling1978) and could be understood in terms of social tipping points with negative outcomes (Juhola et al., Reference Juhola, Filatova, Hochrainer-Stigler, Mechler, Scheffran and Schweizer2022; Spaiser et al., Reference Spaiser, Juhola, Constantino, Guo, Watson, Sillmann, Craparo, Basel, Bruun, Krishnamurthy, Scheffran, Pinho, Okpara, Donges, Bhowmik, Yasseri, Safra de Campos, Cumming, Chenet and Spears2024). We briefly note that such amplifying social dynamics can include human responses to hazards: as one example, trade restrictions in response to global food price shocks usually increase prices further (Alexander et al., Reference Alexander, Arneth, Henry, Maire, Rabin and Rounsevell2023; Clapp & Moseley, Reference Clapp and Moseley2020). In practice, amplification would involve a complicated web of non-linear change within systems as well as interactions between systems. Approaches such as the stress-trigger-crisis model of Lawrence et al. (Reference Lawrence, Homer-Dixon, Janzwood, Rockstöm, Renn and Donges2024) or the earlier causal archetypes of the ‘synchronous failure’ model (Homer-Dixon et al., Reference Homer-Dixon, Walker, Biggs, Crépin, Folke, Lambin, Peterson, Rockström, Scheffer, Steffen and Troell2015) could prove very helpful in understanding this (Section 5.2).

One particularly severe outcome of amplification in the context of GCR could be global societal collapse. While there are many ways to define the latter, one simple, forward-looking option in our case is a rapid development of the global system towards a state where it is no longer able to provide (as it currently does) for the material subsistence of most humans.Footnote ¹¹ It is unclear whether such a collapse could be recovered from (Baum et al., Reference Baum, Armstrong, Ekenstedt, Häggström, Hanson, Kuhlemann, Maas, Miller, Salmela, Sandberg, Sotala, Torres, Turchin and Yampolskiy2019; Belfield, Reference Belfield, Centeno, Callahan, Larcey and Patterson2023). There has been much work on the subject of collapse, most notably in the case of past societies (Brozović, Reference Brozović2023; Centeno et al., Reference Centeno, Callahan, Larcey and Patterson2023; Tainter, Reference Tainter1988) but also in ecology (Cumming & Peterson, Reference Cumming and Peterson2017) and in complex evolutionary systems more generally (Arnscheidt & Rothman, Reference Arnscheidt and Rothman2022). While a wide range of possible mechanisms has been identified, one key consensus matters for our purposes: collapse may be set in motion by a particular hazard, but ultimately plays out due to feedbacks and mechanisms internal to the system (i.e. vulnerability and amplification).

3.3. Vulnerability

In the context of GCR, vulnerability is also emergent from within the global system. For instance, a key point of the literature on emergent global risk is that there is greater potential for amplification in the global system than there used to be (Centeno et al., Reference Centeno, Nag, Patterson, Shaver and Windawi2015; Goldin & Mariathasan, Reference Goldin and Mariathasan2014; Helbing, Reference Helbing2013; Homer-Dixon et al., Reference Homer-Dixon, Walker, Biggs, Crépin, Folke, Lambin, Peterson, Rockström, Scheffer, Steffen and Troell2015; Lawrence et al., Reference Lawrence, Homer-Dixon, Janzwood, Rockstöm, Renn and Donges2024). This increased potential for amplification is one form of increased vulnerability. The severity of possible outcomes may also be worse now than in the past; while the emergence of new and dangerous technologies (i.e. hazards) has played a role in this (Ord, Reference Ord2020; Rees, Reference Rees2004), so too has the fragility and potential irreplaceability of global critical infrastructure (Manheim, Reference Manheim2020). Here, we frame our analysis around three key sources of emerging vulnerability: increased global interconnectedness, decreased global diversity (in a variety of domains, particularly in terms of possible responses to disruptions), and humanity’s reliance on advanced technology as well as complex sociotechnical systems.

The trend of increasing global interconnectedness is readily apparent. For example, global food trade flows have increased substantially in the last few decades (D'Odorico et al., Reference D'Odorico, Davis, Rosa, Carr, Chiarelli, Dell'Angelo, Gephart, MacDonald, Seekell, Suweis and Rulli2018; Puma et al., Reference Puma, Bose, Chon and Cook2015), and the yearly number of air traffic passengers doubled from 2 to 4 billion between 2000 and 2019 (facilitating the potential spread of pandemics; Baker et al., Reference Baker, Mahmud, Miller, Rajeev, Rasambainarivo, Rice, Takahashi, Tatem, Wagner, Wang, Wesolowski and Metcalf2022). Increased interconnectedness in the modern global system is typically identified as a major driver of amplification and global systemic risk (Centeno et al., Reference Centeno, Nag, Patterson, Shaver and Windawi2015; Goldin & Mariathasan, Reference Goldin and Mariathasan2014; Helbing, Reference Helbing2013; Homer-Dixon et al., Reference Homer-Dixon, Walker, Biggs, Crépin, Folke, Lambin, Peterson, Rockström, Scheffer, Steffen and Troell2015; Lawrence et al., Reference Lawrence, Homer-Dixon, Janzwood, Rockstöm, Renn and Donges2024; Sillmann et al., Reference Sillmann, Christensen, Hochrainer-Stigler, Huang-Lachmann, Juhola, Kornhuber, Mahecha, Mechler, Reichstein, Ruane, Schweizer and Williams2022). Interconnectedness decreases a system's susceptibility to smaller shocks, by allowing the flow of resources to make up for localised shortfalls, but increases its susceptibility to larger disruptions, by allowing failures to cascade (Foti et al., Reference Foti, Pauls and Rockmore2013; Helbing, Reference Helbing2013; Scheffer et al., Reference Scheffer, Carpenter, Lenton, Bascompte, Brock, Dakos, Van de Koppel, Van de Leemput, Levin and Van Nes2012; Young et al., Reference Young, Berkhout, Gallopin, Janssen, Ostrom and Van der Leeuw2006). An illustrative thought experiment is the following (Siegenfeld & Bar-Yam, Reference Siegenfeld and Bar-Yam2020): imagine you have 100 ladders leaning up against a wall, and then you tie them all together. Each individual ladder is much less likely to fall, but if they do fall they will all fall at once.

A second important trend affecting vulnerability is a loss of global diversity. This has been occurring in a wide range of contexts, from language to institutions to biology (Williams et al., Reference Williams, Zalasiewicz, Haff, Schwägerl, Barnosky and Ellis2015; Young et al., Reference Young, Berkhout, Gallopin, Janssen, Ostrom and Van der Leeuw2006). One specific instructive example is in the global food system: food production is increasingly reliant on a small number of staple grain species, dominated by a small number of companies, and dominated by a small number of countries (Clapp, Reference Clapp2023; Nyström et al., Reference Nyström, Jouffray, Norström, Crona, Søgaard Jørgensen, Carpenter, Bodin, Galaz and Folke2019). This allows for greatly increased short-term productivity, but makes us more vulnerable should conditions suddenly change: for example, what happens if one of these staple grain species or key global suppliers fails for some reason? More generally, the loss of diversity is an issue because diversity is part of how CASs retain resilience (Folke et al., Reference Folke, Carpenter, Walker, Scheffer, Elmqvist, Gunderson and Holling2004; Levin, Reference Levin1998; Levin et al., Reference Levin, Xepapadeas, Crépin, Norberg, De Zeeuw, Folke, Hughes, Arrow, Barrett, Daily, Ehrlich, Kautsky, Mäler, Polasky, Troell, Vincent and Walker2013). A particularly useful framing is that of response diversity: maintaining a variety of potential response behaviours provides complex systems – including the global system – with the ‘raw material’ for adaptive capacity after disruptions (Walker et al., Reference Walker, Crépin, Nyström, Anderies, Andersson, Elmqvist, Queiroz, Barrett, Bennett, Cardenas, Carpenter, Chapin, de Zeeuw, Fischer, Folke, Levin, Nyborg, Polasky, Segerson and Vincent2023).

A third key trend is humanity's increasing reliance on advanced technology – and indeed, on complex sociotechnical systems (composed of humans and their interactions with technology) that no individual human fully understands. Returning to the food system example, developments in industrial agriculture (breeding new high-yield crop strains, large-scale fertiliser production, and machinery) vastly increased global yields throughout the 20th century (Evenson & Gollin, Reference Evenson and Gollin2003; Smil, Reference Smil2004); yet, most humans now depend on these technologies for their survival (Moersdorf et al., Reference Moersdorf, Rivers, Denkenberger, Breuer and Jehn2023; Smil, Reference Smil2022). Greater integration of AI into agricultural systems (Galaz et al., Reference Galaz, Centeno, Callahan, Causevic, Patterson, Brass, Baum, Farber, Fischer, Garcia, McPhearson, Jimenez, King, Larcey and Levy2021; Tzachor et al., Reference Tzachor, Devare, King, Avin and Ó hÉigeartaigh2022) will amplify this dependence. Technological dependence can interplay with global interconnectedness: as a smaller-scale case study, we can consider the recent shortage of new cars in the United States due to semiconductor shortages on the other side of the globe (Dziczek, Reference Dziczek2022).

In practice, we depend not just on the advanced technologies themselves but also on the increasingly complex sociotechnical systems within which the technologies are manufactured, distributed, and used. Sociotechnical complexity, despite its other benefits, sets us up for hard-to-prevent cascading failures (Perrow, Reference Perrow1999). More critically, increases in complexity are often irreversible, as important infrastructure and knowledge pertaining to older approaches are lost (Manheim, Reference Manheim2020) – increasing the severity of the worst-case outcomes.

Of course, these three trends have brought substantial benefits. Beyond the other benefits of globalisation, increased interconnectedness reduces the risk of smaller disruptions. Reduced diversity and advanced technology both allow for increased productivity and efficiency under a specific set of circumstances and thus plausibly also help the global system buffer against certain smaller shocks. Yet, despite these benefits, we suggest that each of these three trends increase the likelihood and potential severity of global catastrophic outcomes. Interconnectedness allows failure to spread much more quickly to larger scales, and a lack of diversity means the global system will struggle to adapt to certain unexpected disruptions. The dependence on technology means that scenarios involving some loss or failure of this technology lead to much more catastrophic outcomes than they otherwise would.

If our vulnerability in the context of GCR has indeed been increasing, why is this? At one level of explanation, we can highlight economic incentives to prioritise efficiency (short-term reliable productivity) over resilience. This is particularly apparent in ecosystem management, where there is a long history of humans attempting to ‘optimise’ an ecosystem (e.g. for productivity) and later finding, often at great cost, that key elements of resilience were lost in the process (Holling & Meffe, Reference Holling and Meffe1996; Scott, Reference Scott1998). For much the same reasons which lead to insufficient governance of hazard-creating actors (Section 3.1), there is insufficient governance of vulnerability-creating actors. As long as these actors (individuals, companies, or states) can obtain a short-term gain from actions that increase global vulnerability (prioritising efficiency over resilience), they will do so.

Yet, things are also more complicated. Many of the key factors driving global vulnerability – including the three trends outlined earlier, are ultimately emerging in a complex decentralised manner. As one additional example, the fact that our complex global infrastructure can develop highly connected hubs that propagate disruptions (e.g. Mani et al., Reference Mani, Tzachor and Cole2021) is due to fundamental aspects of bottom-up network development, such as preferential attachment (Newman, Reference Newman2018), which are very difficult to avoid. Nevertheless, governments and global institutions can proactively invest in resilience, for example, by investing in and promoting response diversity (Walker et al., Reference Walker, Crépin, Nyström, Anderies, Andersson, Elmqvist, Queiroz, Barrett, Bennett, Cardenas, Carpenter, Chapin, de Zeeuw, Fischer, Folke, Levin, Nyborg, Polasky, Segerson and Vincent2023), actively maintaining backups for key critical infrastructure systems, carefully modularising those systems (see Tzachor et al. (Reference Tzachor, Richards and Holt2021) as one example), and so on.

Ultimately, much like the emergent generation of hazards, the generation of vulnerability is a deep consequence of the fact that selection in CASs occurs at lower levels (Levin, Reference Levin1998; Levin et al., Reference Levin, Xepapadeas, Crépin, Norberg, De Zeeuw, Folke, Hughes, Arrow, Barrett, Daily, Ehrlich, Kautsky, Mäler, Polasky, Troell, Vincent and Walker2013) and is thus not necessarily beneficial to the system as a whole. The generation of vulnerability in the global system can also be related to the growth phase of the adaptive cycle (Homer-Dixon, Reference Homer-Dixon2006): this involves precisely increasing connectivity and decreasing diversity.

3.4. Latent risk

Another, perhaps more easily overlooked, systemic contribution to GCR involves latent risk. Latent risk refers to risk that is dormant under one set of conditions but becomes active under another (Kemp, Reference Kemp2021; Kemp et al., Reference Kemp, Xu, Depledge, Ebi, Gibbins, Kohler, Rockström, Scheffer, Schellnhuber, Steffen and Lenton2022). One illustrative example of latent risk on a global scale is the following (Kemp, Reference Kemp2021): while stratospheric aerosol injection (SAI) could cool the planet and potentially reduce global warming, it also introduces the risk of ‘termination shock’, in which the planet warms very quickly if aerosol injection were to suddenly stop (Parker & Irvine, Reference Parker and Irvine2018). Here, the risk is created by the introduction of SAI, but remains latent at first – it is only activated if SAI suddenly stops.

Latent risk can also be thought of as reflecting path dependence within the global system. Path dependence means that the evolution of the system depends not just on its present state but also on its history – in other words, the path taken to get there – and is a fundamental property of CASs (Levin, Reference Levin1998). The idea of latent risk highlights that the global system may evolve along different ‘paths’ in the future, that on some of these paths humanity is exposed to a much greater risk of catastrophic outcomes than on others, and, critically, that events and processes taking place in the present can affect the magnitude of the risk faced in these future system states. While one could try to conceptualise this behaviour purely in terms of the interaction of hazards and vulnerabilities over time, we find the concept of latent risk a particularly powerful one and thus include it as the fourth component of our conceptual framework.

The global system is actively generating latent risk. For example, beyond its immediate impacts, climate change could hamper humanity's ability to recover from other catastrophes (Kemp et al., Reference Kemp, Xu, Depledge, Ebi, Gibbins, Kohler, Rockström, Scheffer, Schellnhuber, Steffen and Lenton2022). More generally, if any of the global critical systems on which humanity depends on were to fail (Avin et al., Reference Avin, Wintle, Weitzdörfer, Ó hÉigeartaigh, Sutherland and Rees2018), this might trap us in a state where that system could not be regenerated, and would also decrease our species’ resilience to further catastrophes, such as human extinction. If increasing sociotechnical complexity leads to the loss of simpler alternatives (Manheim, Reference Manheim2020; see also Section 3.3), this present-day trend increases the latent risk in such scenarios. To illustrate these issues while maintaining a focus on the worst-case outcomes, we now consider how systemic risk would manifest after global societal collapse (noting that any reasoning about post-collapse worlds is necessarily speculative).

After a collapse, emergent phenomena might trap the system in the collapsed state. Various theories of human civilisational development – i.e. the development of the global system – emphasise the role of amplifying feedbacks. For example, the industrial revolution may have been substantially driven by feedback cycles that rapidly increased human access to energy (Lenton & Scheffer, Reference Lenton and Scheffer2024). However, if the global system were to collapse and advanced technology were to be lost, it is not clear that this pattern could be repeated, because most easily accessible fossil fuels would have been used up (Baum et al., Reference Baum, Armstrong, Ekenstedt, Häggström, Hanson, Kuhlemann, Maas, Miller, Salmela, Sandberg, Sotala, Torres, Turchin and Yampolskiy2019; Belfield, Reference Belfield, Centeno, Callahan, Larcey and Patterson2023).Footnote ¹² If amplifying feedbacks initially propelled the global system to its present state, the lack of such feedbacks – and thus, a dominance of stabilising feedbacks – could prevent re-industrialisation after a collapse.

After a collapse, emergent phenomena could also substantially increase the risk of human extinction. Without industrial methods of food production, the global population would fall to a substantially lower level and may also become geographically disconnected. Surviving human populations could then face non-linear ecological dynamics akin to those faced by non-human species (May, Reference May1977; Scheffer et al., Reference Scheffer, Carpenter, Foley, Folke and Walker2001), including runaway extinction if population numbers fall below a minimum size.Footnote ¹³ Crucially, minimum viable population sizes are often set by the influence of population size on the ability to effectively cooperate (Stephens & Sutherland, Reference Stephens and Sutherland1999) and may thus be much higher than those predicted purely on the basis of genetics (Baum et al., Reference Baum, Armstrong, Ekenstedt, Häggström, Hanson, Kuhlemann, Maas, Miller, Salmela, Sandberg, Sotala, Torres, Turchin and Yampolskiy2019). While complete human extinction remains a high bar, one key point is worth emphasising: because of latent risk and emergent phenomena, a given catastrophe does not need to make Earth completely uninhabitable to ultimately lead to human extinction.