By Craig Segall & Ben Margetts
Major earth systems are now on the brink of irreversible change as a result of climate change, in ways that will substantially increase near-term human suffering and long-term risk, but almost no government capacity is dedicated to interventions that could stabilize these systems. Climate policy instead focuses almost entirely on emissions reductions and adaptation; those are important steps but, on their own, cannot avoid cascading risks that will cause trillions in economic harm, threaten tens of millions of lives, and divert government capacity that could be used to address existential risks. We know we're new here -- but are sharing this analysis because we think these issues are genuinely important, and look forward to engaging on these draft arguments.
As senior policy practitioners, we judge government capacity to stabilize earth systems (“climate stabilization”) to be a highly neglected domain of work in which marginal investments on the order of millions of dollars could – if applied soon – leverage risk reductions valued in the trillions. This is a pivotal moment for these investments, as system risk is rapidly rising, and governments must act soon, or face unstoppable planet-scale disruption. In light of this analysis, we have pivoted to address these emerging issues after careers working on emissions mitigation. We are writing this to urge attention to this entire category of climate response, which has been almost entirely overlooked.
The effective altruist (EA) approach to cause prioritization is a useful lens for this argument. We have considered the foundational EA writing on climate – Halstead's report, Ord's treatment in The Precipice, the 80K problem profile, and other analyses. These pieces generally concur that climate change risks are significant, but generally view them as not neglected, and under-prioritize stabilization government capacity investments. We found merit in those analyses when they were written. But the evidence base, risk calculus, and dynamics of the problem have shifted since then, in ways that point toward a different set of conclusions largely because it is now clear that system risk is higher than anticipated, and the likelihood of mitigation reducing these risks acceptably much lower. We set out our cruxes explicitly at the end and welcome pushback, especially on the quantitative claims.
Conflict of Interest: We are managers at Climate Hub, a small policy advocacy nonprofit on these issues. Our bias matches our argument. We advocate for the research, development, and governance of stabilization options - not for their deployment. We think this is an area that warrants attention well beyond our specific effort.
Acknowledgments: Thanks to many readers who have sharpened this argument, including the team at Rethink Priorities who encouraged us to dig deep into the scientific literature, and which is doing research and building models on these issues as we write. Thanks too to Ben Rachbach, who pointed out interactions with AI safety, Liam St Louis who sharpened our discussion of timing and cascading risk, David Bowes who articulated ways neglecting these issues could derail management of other risks, Joshua Elliot who aimed our attention at catastrophic nonlinear climate risk, Adam Goff who showed gaps in the analysis of discount rates and catastrophic risk, and Taren Stinebrickner–Kauffman who helped sharpen and focus the entire argument. Opinions and errors, of course, are our own
LLM Disclosure: We drafted this post ourselves, but asked an LLM to review the arguments after a few rounds of drafting, and incorporated some suggestions. We've since rewritten it extensively -- based not just on that review but on a great deal of human feedback -- and endorse it.
TL;DR
Climate stabilization – the research, monitoring, and potential use of tools to manage catastrophic earth-system risk – is a neglected global catastrophic risk (GCR) intervention that belongs in the EA catastrophic-risk portfolio. The field receives under $15 million per year globally for governance and advocacy, against expected risks in the trillions – a funding profile that sits in the same ballpark as other EA-adjacent governance areas before they reached their current scale. We argue as follows:
We are locked into a climate trajectory that, the evidence now shows, puts the world firmly at risk of earth system changes – like new research showing we are on track for the collapse of major ocean currents that could drastically cool northwestern Europe and disrupt monsoon rains supporting agriculture for billions, and the irreversible melting of ice sheets that could spike global sea levels and trigger trillions in damages and adaptation costs. New evidence shows that emissions mitigation cannot, alone, prevent these risks because emissions cannot fall fast enough to offset warming as Earth’s “radiative imbalance” – the net energy it absorbs – hits record levels. That is both because economic systems cannot deliver sharp enough cuts and because Earth is also rapidly getting darker and absorbing more energy as it does due to loss of reflectivity, in significant part because of warming–linked changes to clouds and to melting ice, along with cuts to aerosol pollution – meaning that there are multiple factors driving us towards tipping points. Once breached, these tipping points will impose trillions in costs and threaten the lives of millions and the welfare of billions – which will also reduce the ability of political and economic systems to address other critical risks.
Three categories of resulting risk are germane here, each independently sufficient for upping government responsive capacity:
These serious risks, including the risk of derailing responsive capacity to other existential risks, each warrant a new look at climate change. Mainstream EA analysis somewhat deprioritizes climate work because so many organizations in emissions reduction are already engaged, arguably limiting the value of marginal work. But that analysis is primarily for emissions mitigation – not for alternate approaches that are profoundly neglected. Climate stabilization – the approach we are arguing needs to be added to the portfolio – has large potential to reduce climate risks in ways that save millions of lives and free up trillions of dollars – but thus far has received only millions in funding. These are neglected solutions to a non–neglected problem, and the governance window for shaping them responsibly and effectively is closing.
We are making an option value argument, at core, along all three of the risk pathways we have just outlined. Stabilization technologies – like stabilizing ice sheets to slow catastrophic sea level rise, or reflecting a small percentage of sunlight back to space to reduce the earth’s energy imbalance – can likely operate for tens of billions of dollars annually to reduce trillions in risks that would otherwise be incurred. That would provide time for emissions mitigation, and greenhouse gas removal from the atmosphere, to fully stabilize the climate – essentially getting humanity through a major bottleneck. We think the option value of developing and governing these approaches now is high; investments of tens to hundreds of millions in policy advocacy and research could mature these options and support governance. Leveraging trillions in risk reduction for millions in initial spending (and with long–term operations costs if these approaches are deployed, still less than ~5% of the cost of avoided risks) is a sensible investment.
We think these options are particularly worth purchasing if we do not improperly discount the future. Far too many governments and standard economic assessments discount future harms. Even at those discount rates, these harms are immense. But if you properly value future lives – even lives as close as the mid-century, when earth system collapse will be well underway – increasing spending on options that could avoid collapse becomes all the more important.
Because of the growing risk of earth system destabilization, and rapidly developing politics, the window to invest in the development and governance of these technologies is open now. We argue that:
There are currently two main climate strategies – mitigation and adaptation. Mitigation aims to reduce human emissions of greenhouse gases; adaptation protects communities from the physical effects of climate change. Both are urgent and essential. These are the solutions that core EA writing on climate largely considers.
Earth’s temperature is fundamentally caused by an energy imbalance – one driven both by rising emissions and diminishing ability to reflect in-coming energy to space; this imbalance is at its highest point ever. As rising temperatures threaten to destabilize critical climate systems, we are confronting risks that are already materializing. Mitigation alone cannot reliably avert these risks and adaptation cannot fully protect us from them. A third strategy is now required.
Climate stabilization is the research, monitoring, and management of severe earth system destabilization risk – it aims to prevent catastrophic instability or the crossing of thresholds that lead to irreversible trajectories of change.
Mitigation, adaptation and stabilization work together - stabilization is not a substitute for the other two but the element that protects them both.
Climate disruption at the scale overshoot also erodes the political and economic conditions on which a successful energy transition depends. Nonlinear disruptions destroy government capacity and funding that could push forward the transition – and manage other risks. Huge costs from wildfires, sea level rise, and similar disasters dry up public budgets and time. And their consequences matter too: Major adaptation efforts to address these disasters also cost billions or trillions, and political consequences – like authoritarian responses to mass climate migration or to budget pressures – could further destabilize the very global supply chains that the transition from fossil fuels to solar +battery power relies on. Stabilization is what keeps the transition viable.
Stabilization targets the hazard–generating systems themselves. Several of these systems are already showing signs of destabilization, including:
Managing these systems requires research, monitoring, governance, and response capacity that the existing climate architecture – built to manage emissions, not to monitor or intervene in earth systems directly – was not designed to provide. The field includes:
Glacier stabilization would involve engineering approaches to slow the retreat of major ice sheets – proposed methods include underwater sills to block warm water intrusion beneath glaciers, subglacial drainage to remove meltwater that lubricates ice flow, and targeted cooling of ice sheet bases. Solar radiation management (SRM) would involve spraying small reflective particles to reflect a percent or so of incoming sunlight back to space; that could be done by planes flying in the stratosphere, by ships spraying reflective particles to brighten clouds, or similar means. Both are discussed in more detail in the tractability section below.
Durable stabilization also has a longer horizon. Even where near–term interventions succeed in reducing hazard intensity, lasting stability ultimately requires returning greenhouse gas concentrations to safe levels – which will require carbon removal at significant scale. That work must begin now to be meaningful later.
The EA community's dominant position on climate, articulated by Toby Ord in The Precipice, John Halstead's supplementary report to What We Owe the Future, and 80,000 Hours' problem profile, holds that direct existential risk from climate change is lower than from AI or engineered pandemics, but that climate change is still a very important problem. Ord's estimate of approximately 1–in–1,000 existential risk from climate by 2100 is lower than his estimates for unaligned AI (1–in–10) or engineered pandemics (1–in–30).
We are not arguing that climate stabilization should displace those priorities (we after all are not expert on them). We are arguing that climate risks are serious in their own right, may displace government ability to solve other problems some EAs prioritize, and that stabilization solutions need considering more deeply.
This argument rests on the three risk pathways we describe at the outset – all of which focus not on how the climate responds to emissions, but on how human societies respond to earth system collapse. Standard EA analysis asks: how much warming will we get? That question is at least improving in clarity – with the significant caveat that the new evidence we cite above on global darkening may turn out to accelerate warming after all.
But social systems matter too. The more important question, now that we are in overshoot, is different: at what point does warming overwhelm society's capacity to cope? On that question, we know remarkably little – with earth system and social system uncertainties magnifying each other. Even modest assumptions about this threshold change the risk picture dramatically. A 2% probability that 3°C of warming is sufficient to trigger civilizational collapse – combined with roughly 50% probability of reaching 3°C on current trajectories – approximately doubles the total existential risk estimate. This matters because Halstead's analysis examined each risk in isolation – heat, impact on food systems, sea level, conflict – and essentially assumed that society's capacity to adapt would scale with the damage. That may hold when systems fail one at a time. It is much less likely to hold under overshoot conditions, where ice sheets, ocean currents, ecosystems, and carbon feedbacks destabilize simultaneously – compounding stress on food, financial, and political systems in ways that a one–dimensional risk analysis can't capture.
The climate risk landscape in 2026 is not what it was when EA's foundational analyses were written. It has become clear that the risk from global warming is intensifying, along multiple causal pathways, and that the assumed solution – emissions reductions – are not sufficient on their own. That combination of factors make climate stabilization critical to explore, as it addresses these risks far more completely.
We should think about the risk in terms of Earth's energy imbalance, not just emissions curves. NASA's CERES satellite program measures how much solar radiation Earth absorbs (ASR) and how much heat escapes to space (OLR). The spiking imbalance is a surprise – extending beyond many model predictions. Since 2000, ASR has increased by 1.7 W/m². This means Earth is taking in substantially more energy than before: energy that accumulates mostly in the oceans, driving continued warming, thermal sea–level rise, and more energetic climate systems. The main suspected driver is changes in cloud cover. Some clouds reflect incoming sunlight, others trap outgoing heat. We are gaining heat trapping clouds and losing cooling clouds.
To make that number concrete: an ASR increase of 1.7 W/m² is equivalent in forcing terms to increasing atmospheric CO₂ by approximately 138 ppm – on the order of the entire observed increase from direct emissions since the Industrial Revolution.These are order–of–magnitude comparisons, not precise equivalences, but the scale is striking.
Earth's Energy Imbalance (EEI) – the net gap between energy coming in and heat going out – has increased by approximately 1 W/m² since 2001, nearly twice the increase predicted by climate models over that period. This is its highest level ever. A sustained imbalance at this level implies approximately 0.75°C of additional long–run warming even under a net–zero emissions scenario, because temperatures keep rising as long as incoming energy exceeds outgoing energy.
This is not a model projection but a satellite observation - and only recently have its implications for reflectivity as an independent driver of risk become clear. It changes the structure of the risk analysis: we are not only discussing what might happen under various emissions scenarios, but what is already happening in the earth's energy budget. We have both global warming and global darkening – and those two trends are reinforcing each other. The planet is getting much less reflective – as the result of melting ice, reduction in aerosol pollution and likely also as the result of warming–linked changes to cloud cover among other factors – which means it is suddenly absorbing even more energy:
Confronted with new systems of risks, we should look for new collections of solutions. This focus helps identify neglected solutions outside of emissions mitigation, because emissions are just one lever on how much energy Earth absorbs. Reflecting sunlight is another lever – and we can also respond to the consequences of the imbalance directly, for instance by intervening into collapsing ice sheet systems with engineering solutions.
The current warming demonstrates how important it is to think along multiple pathways at once – because the warming spike in the last few years is driven in substantial part by the planet becoming less reflective.
Here’s a direct example: Ships used to have high–sulfur emissions – posing a real health risk to port cities, but reflecting lots of sunlight to space, including by forming marine clouds. When the International Maritime Organization (IMO) mandated sulfur reductions from ships in 2020, it inadvertently terminated a large–scale accidental geoengineering program. Marine vessels had been seeding reflective low–level clouds over shipping lanes for decades. Removing that sulfur produced a measurable warming signal, in a mini–termination shock. This is a real, observed climate response to a specific human action, demonstrating that anthropogenic albedo modification works in both directions at scale.
This analysis demonstrates why the global focus on emissions reductions, alone, is insufficient – and hence why stabilization is its own important focus. This figure from the Intergovernmental Panel on Climate Change’s most recent (2023) synthesis report shows how implausible a mitigation only pathway has become.
There is no plausible political or economic pathway to reach warming below 2 degrees by 2100 without dramatic shifts in global policy that do not look at all likely (extremely unlikely in our view, given major retreats on climate policy from many major emitters, including the US and EU that are removing key legal supports for mitigation and canceling existing public investments, combined with complex public policy challenges associated with swapping out global energy systems).
To be clear, we’ve made real progress and we think a long–term path to a low–carbon world is likely – but that path is not going to be quick enough to avoid major risks.
There are significant risks associated with warming in the range of warming we are experiencing. These include major threats to human health and the economy, and damage to major ecosystems and earth systems. Critically, these risks include a host of nonlinear, or tipping point, risks that, once triggered, cannot readily be reversed – and all of which become much more probable at the warming levels we will be seeing. The collapse of major ice sheets is now likely in the view of the IPCC, and we are seeing concerning reports that collapse of the Gulf Stream within a century is possible (with over 70% of model runs with high emissions trajectories in a recent study leading to collapse). These collapses may in turn drive major further release of greenhouse gases, and will also impose huge stresses on public resources and capacity.
In fact, researchers judge we are moving into zones of risk for many of these systems:
We do not see a path to simply “adapt” to risks like the collapse of major earth systems without major social and political costs that look very unlikely to be available (high confidence given the multi–trillion dollar cost of, for example, global sea walls to address sea level rise, or the enormous political and economic complexities associated with dropping European temperatures to Arctic levels if the Gulf Stream collapses).
That means climate stabilization is important, even given continuing focus on mitigation.
Stabilization also matters for two reasons which prior EA analyses have not adequately addressed, particularly the second.
Direct welfare case. McKinsey estimates funds needed over $200 trillion ($9.2 trillion in annual spending, over the next 24 years) to reach global net zero by 2050 through mitigation; the International Energy Agency estimates $120 trillion.
We do not think there is even a moderate chance that the world mobilizes funds at that scale by 2050 (very unlikely, given the persistent failure of the world’s governments, even under treaty obligations, to fund existing commitments to mitigate loss and damage from climate harms to Global South nations). Those costs, in the absence of climate stabilization, would be borne at the same time as climate–linked disasters intensify; climate-linked disaster costs alone already exceed $115 billion for the U.S. in 2025. So the likely trajectory will be escalating human suffering and growing costs to government, which will compete with spending that could go to the transition and to other social goods.
Those costs will also be felt as increases to the cost of living – and we are seeing that politicians often respond to affordability concerns, perversely, by cutting back on climate ambition because of the near–term costs of those policies despite their long–term benefits. As MacAskill observes in What We Owe to the Future, "[t]here is a substantial chance that our decarbonisation efforts will get stuck" – a risk he puts at "about one in three." We agree.
The IPCC projects that under moderate–to–high warming, climate change will cause hundreds of thousands to over a million additional deaths annually by 2100 through heat stress, malnutrition, disease, and displacement. Cumulated across decades and weighted for morbidity as well as mortality, this implies hundreds of millions to over a billion disability–adjusted life years (DALYs) at stake across the century. The relevant question for climate stabilization specifically is what fraction of that cost well–designed interventions could plausibly avert, and at what cost. GiveWell's top charities avert a lost DALY at roughly $3,000–$5,000. To match that benchmark with $100 million in climate stabilization governance investment, you'd need to avert approximately 20,000–33,000 DALYs. Given a welfare cost base running into hundreds of millions of DALYs, that requires the governance investment to affect roughly 0.002–0.003% of expected harm – an extremely low bar for serious senior–level advocacy work to clear.
Further, each major climate tipping point inflates risk, heightening nonlinear risks, and increasing the value of acting to avoid these cascades. The Thwaites glacier in Antarctica, on its own, could raise sea level by half a meter, submerging coastal cities or imposing enormous adaptation costs. Approximately 230 million people currently live below 1 meter above current high tide lines. And sea level could raise by more than a meter: Modelling by Kulp and Strauss (2019) found up to 630 million people living on land below projected annual coastal flood levels by the end of the century under high emissions. Because the cost of engineering solutions to slow Thwaites is likely in the tens of billions while the costs of sea level rise are easily in the trillions, detailed feasibility studies make sense – and this is just one of several systems at risk.
Indirect importance: compound existential risk. This is the argument that has received the least rigorous treatment relative to its potential importance. Climate overshoot degrades government capacity to manage other potential risks, including existential risks. Even a small increase in the probability of catastrophic outcomes as a result – for example, less institutional bandwidth for pandemic preparedness, nuclear security, or AI governance – produces importance scores that dwarf direct welfare calculations. Under standard longtermist reasoning, a 0.1 percentage point increase in existential risk probability is equivalent in expected value terms to the loss of billions of future lives. But one need not look just to the far future – compounding risk from government capacity losses can act quickly to cause harm, and to derail a host of important efforts.
Major disasters strain governments, which almost invariably are acting with limited budgets and staffing. The recent Los Angeles wildfires, for example, create multi–year, multi–billion dollar losses for the world’s fourth largest economy’s public budget. These losses are not easily recouped; governments generally do not have political license or legal rein to simply raise taxes to match these costs, and the private sector is also facing major constraints, with the collapse of insurance markets under climate risk as a primary example. Each notch of increased climate risk corresponds to greater budget pressure, and hence greater pressure on limited government staffing. These strains already substantially limit government responsive capacity - and compound sharply under earth system failure scenarios.
As a key example, mass displacement at climate–overshoot scale generates sustained political crises, hardens border regimes, and erodes the international cooperation on which long–horizon risk management depends. It also creates the conditions – overwhelmed health systems, disrupted ecosystems, species and people pushed into novel contact – in which new pathogens emerge and spread most readily. Carlson et al. (2022) projected approximately 4,000 novel cross–species viral transmission events by 2070 from climate–driven species range shifts alone. Mora et al. found that 58% of known human infectious diseases have already been aggravated by at least one climatic hazard. The pandemic pathway is particularly significant from an EA perspective: it is mechanistically clear, quantified in the peer–reviewed literature, and connects directly to biosecurity risks the community already prioritises. And it is not the only such pathway – as the ongoing fallout from the covid pandemic in global politics shows, major disasters have complex and often negative effects on long–term political capacity.
The governance capacity loss that results – fiscal resources diverted to disaster response, political attention consumed by crisis management, multilateral cooperation under strain – directly reduces the bandwidth available for AI governance, biosecurity, and other long–horizon existential risk work. Governance capacity is finite, great–power cooperation is fragile, and institutions managing cascading climate crises have less bandwidth for the technically complex, long–horizon work that catastrophic risk management requires.
We don't know the exact magnitude of these interactions, and they are rarely analyzed in depth, including in EA literature – but that gap should not obscure the reality of rising risk or the value of marginal reductions in it. Even a small probability that climate overshoot materially increases existential risk through these mechanisms would, under standard longtermist assumptions, place climate stabilization high on priority lists
The bottom line from our perspective? Global net zero is already off–track, and it is at risk of being further derailed by the pressures that climate change itself creates on government and financial systems. This derailment risk is serious, and worsens the bottleneck. And if the transition derails, the many human and system risks caused by overshoot further intensify, in worsening physical and social feedback loops. It is hard to see a path forward to avoid these risks that does not include at least exploring climate stabilization. The EA community correctly recognizes that climate policy is all about managing risk, not just managing emissions.
EA's prior analyses are right that climate change as a whole is not neglected – global climate finance exceeded $1 trillion annually as of 2024. The standard EA argument for deprioritising climate giving rests substantially on this observation, and it is correct as applied to mitigation. But our argument is that mitigation, alone, is insufficient – and that stabilization is a highly neglected response to growing risk that should be included in the array of responses. The most recent EA work has also argued for building a climate policy portfolio that is highly robust to political and economic risk, and which can mitigate human risk via multiple means. Climate stabilization should find a place in any such portfolio – but the EA field has not yet followed it there.
EA has not entirely ignored stabilization. Coefficient Giving previously recommended grants for SRM research and governance, including major grants to Harvard's Solar Geoengineering Research Program and the Degrees Initiative for Global South research capacity. That was the right instinct. But Coefficient has not recommended recent grants, and no comparable funder has stepped in. The field is therefore more neglected now than it was at the point of peak EA engagement – and the window to shape early effective governance has continued to close in the interim.
Climate change as a whole is not neglected, but climate stabilization solutions are hugely neglected despite their potential – and more so now than when EA last looked seriously at the space. The neglectedness analysis has simply not been updated to reflect either the changed earth system evidence or the withdrawal of the funders who had previously engaged.
Total global funding for solar radiation management research, glacier stabilization, and related earth–system interventions is approximately $15–$50 million annually (the broad range here is because of differences in how one might count – direct SRM funding, for instance, is about $30 million into the public sector in a given year, but funding of allied science can inflate this figure, as can funding into allied fields). A recent venture capital investment into a private firm exploring SRM of about $60m grows this figure, but also underlines how important it is for government to catch up with private investment to ensure proper risk management. Governance and advocacy work is clearly below $15 million per year based on our review of known grants, with fewer than a few hundred researchers worldwide.
These funding levels are comparable to AI safety funding in the early 2010s, before it became a major EA priority. Yet we now know with certainty – from satellite observation, not model projection – that large risks from climate overshoot are already baked in. Spending on somewhat less certain but plausible risks, like pandemics, stands in the billions.
A simple risk analysis shows that marginal increases in these investments would unlock paths to literally trillions of dollars in reduced risks. Moreover, the barriers to success are known and surmountable. These are sophisticated but well–defined engineering challenges. Governance challenges are harder to define, but it is clear to us that a ramp up in government capacity to govern research and potential deployment would substantially reduce overall risk, given that current capacity is very weak, meaning marginal increases substantially improve outcomes from a low base. So marginal funding can translate directly into measurable progress.
The funding gap is most acute in governance and government capacity. The vast majority of funds in the field go to research. There is almost no funding – less than $15 million from all sources – dedicated specifically to moving governments to engage on these issues, even though it is governments alone that have the capacity to deploy stabilization interventions at a meaningful scale, and even though governance deliberation now is urgently needed to avoid last–minute crisis deployments of stabilization technologies without sufficient testing. To our knowledge, based on an 18–month field survey, fewer than 100 people worldwide are working in advocacy on this issue. Climate Hub, where we work, has a staff of less than 15 and is among the largest organizations working on the politics of this issue globally.
EAs and aligned thinkers appear to us particularly likely to be able to unlock these opportunities because of their thoughtful approach to managed risk. Yet there is also a live institutional tension within EA worth naming. Giving Green added SRM governance as a recommended strategy in 2024. Yet 80,000 Hours' problem profile on climate continues to recommend against careers in solar geoengineering – a position that shapes talent flows more than any single grant decision. The 80K concern rests on a termination shock argument whose primary cited source, Halstead, in fact concludes the risk is overstated. We would like to see 80K either update their profile in light of the evidence reviewed here, or explain what evidence would be required to change their position.
Even relatively small amounts of additional effort on climate stabilization would produce significant gains. Small governance-focused staffs of senior experts can mobilise significant government interest, and unlock private foundation investments in research, for less than $100 million annually – and likely much less to have significant impact. The opportunity to reduce trillions in risk for millions in initial investment is a rare one.
A stray paragraph in the 80,000 Hours profile suggests that solar geoengineering may produce untenable risks of its own, largely from the risk that it may be turned off before emissions have been reduced – a termination shock concern. The profile cites no source that performs an actual risk–risk comparison between managed SRM and unmanaged climate overshoot. The paper it cites – Halstead – in fact concludes this risk is overstated, and that the EA community should likely support research, as other EAs have noted. More recent EA investigations largely recommend continued work on governance. Toby Ord argued in The Precipice that geoengineering "needs to be very carefully governed" but "may well have a useful role to play as an alternative to emissions reductions," adding that using it as a last resort "could lower overall existential risk even if the technique is more risky than climate change itself." Since that time, it has become clear both that climate risks are larger, and that stabilization risks are potentially lower than expected.
The argument for tractability is an option value one. We are not advocating for deploying these technologies – not now, and maybe not ever, if research doesn't bear initial promise out, and if credible governance mechanisms aren’t in place. We are arguing that these options should be fully developed and deployable in the next decade given how serious the coming risks are. If we do not begin work now, these options will be unavailable, at least as well governed possibilities with adequate testing and safety, at the time of highest climate overshoot risk.
We highlight two high value options here – they don’t form an exclusive set, as there are likely multiple paths available, but do illustrate the point. The potential for these options to reduce overshoot risk is high:
Glacier stabilization. The primary mechanism accelerating glacier instability is warm water reaching the underside of ice sheets. Proposed interventions include underwater sills to block warm water intrusion, subglacial drainage to remove meltwater that lubricates ice flow, and targeted cooling of ice sheet bases. These are substantial engineering challenges, but ones with identified physical mechanisms and plausible implementation pathways. Because most sea level rise comes from only a few glaciers, focused civil engineering is a real possibility to avert very large global costs. Multiple organisations are already beginning work, but have raised only millions for initial studies – well behind the amount needed for a substantial effort. A rigorous engineering feasibility study for the full suite of glacier stabilization options doesn't currently exist. Funding one would cost in the range of tens of millions of dollars. If the interventions prove infeasible we lose relatively little; if they prove feasible we'll have dramatically expanded the available options for managing one of the most concrete near–term risks. This is a strong "giving to learn" opportunity.
Solar radiation management management (SRM) would, in principle, work to cool the planet, and remarkably cheaply. Its risks need managing and government capacity needs to scale – but the option is compelling to truly fully vet given the risks potentially avoided. Direct cost estimates are generally in the low tens of billions of dollars per degree – far less than 1% of the federal budget. The infrastructure hurdles are manageable; for planes, for instance, a stratospheric version of effort would involve building or repurposing a fleet on the order of a commercial airline and several airports – a big project, but well within our capacity. The same is true for a shipborne approach: Marine cloud brightening – essentially spraying particles from ships that can create and brighten clouds – is another complementary technological path that is well within global capacity. Once started, SRM would rapidly start to reduce tipping point and nonlinear risks – with sea ice recovering and other polar risks reducing quickly, for instance.
The stratospheric cooling approach would work on the basis of current data, though further safety assessment is of course important: Climate models suggest approximately 1 million tons of sulfur dioxide (SO₂) in the stratosphere annually yields roughly 0.4–1 W/m² of counterbalancing radiative forcing. Counterbalancing the observed approximately 1 W/m² EEI increase would require approximately 1–2.5 million tons of stratospheric SO₂ per year. Industrial processes currently emit roughly 80 million tons of SO₂ annually in the lower atmosphere, so approximately 1–3% of today's industrial SO₂ output, delivered to the stratosphere, would be in the range needed. This grounds SRM in observed data rather than model projections alone. It is increasingly clear that well–designed SRM deployment would disrupt global climate and weather patterns far less than current warming would. The quantity of particles potentially released is small relative to global pollution, and would not have major health impacts. Research and testing is also possible – SRM research can be conducted without risking "termination shocks" that would affect global climate if stopped, and researchers are defining and debating clear engineering parameters for safety.
The shipping example is similarly tractable. Indeed, as we noted above, ship exhaust has actually been cooling the planet for years – and has just recently stopped as a result of public–health–driven changes in marine fuel. Implementing cloud brightness changes over 5% of the ocean via nozzles on ships could cut a degree of warming. More research is doubtless needed, as cooling needs to be designed properly to avoid unbalancing weather patterns. But baseline efficacy is clearly established, and costs are low – in the billions of dollars range.
SRM needs to be properly governed as part of a suite of solutions. It does not work on its own indefinitely if emissions continue to rise, and it does not remove emissions from the atmosphere. It needs to be paired with emissions cuts and with accelerating greenhouse gas removals from the atmosphere to get us all the way through the bottleneck. Further analysis of proposed approaches is also critical, to work through any unanticipated consequences and test out safety and efficacy in detail. This core need implies a strong requirement to develop ethical norms and governance frameworks.
Summing up SRM, we'll emphasize that its clear technical capacity to buy time for the transition is drawing attention – and worthy of EA interest. Political plausibility is also looking real: very recently, we've started to see groups across the political spectrum, as diverse as the Natural Resources Defense Council and the Texas Conservation Coalition Research Institute start calling for accelerated research attention. A private company, Stardust, has also drawn at least $60m in investor interest to develop particles and deployment tools that could be used for SRM by a responsible government. This is a rapidly emerging policy area, but it remains funded at levels far below levels needed to scale research quickly, much less to deploy – consistently in the tens of millions only, very little of that from governments aside from a recent UK–government–funded research program.
On the moral hazard objection. The most persistent barrier to EA engagement with SRM research is the concern – rooted in theory rather than in practice – that making climate stabilization interventions seem feasible reduces political will for emissions cuts. The empirical evidence for this effect is quite weak. Studies to date show no clear direction – depending on how SRM is presented, governed, and discussed, survey-based studies often show no change in attitudes towards mitigation at all, or even increased focus on mitigation in tandem with SRM. The literature overall is mixed; we don't want to overstate this. But large–scale empirical work consistently fails to find the effect that theory predicts.
There are also good reasons rooted in the logic and foundational economics of the energy transition underway to think that moral hazard risks are more nuanced and limited than what is usually portrayed. These reasons include:
On the termination shock objection. A second major objection is that SRM activities would have to be continued once started at global scale, or risk a “termination shock” in which global temperatures suddenly spike. We agree with Halsted that this risk, though serious, is not a reason to avoid governance and research. There are several reasons for this:
On the governance objection. A collection of academics and generally left–aligned advocates have argued for “non–use” of SRM. The general argument is that it is very difficult to govern global technologies. We agree that governance is difficult. However, we observe that the mere fact that a technology is difficult to govern does not mean that it will not be developed (as AI, defense tech, biotech, and so on show). Also, humans have proven able to govern, however imperfectly, many potentially dangerous technologies, including, to date, nuclear weapons. Governance difficulties, in our view, are a reason to scale up research and governance capacity, not to pull funds and hope no one develops the technology anyway.
SRM is technologically feasible enough that it will be seriously considered by governments facing severe climate impacts regardless of whether EA funds research. The real choice is between robust governance frameworks that allow an informed, coordinated decision if deployment is seriously considered, or decisions made in a crisis without frameworks, potentially unilaterally by actors with the most to lose. Governance–focused investment specifically reduces the risk of ungoverned deployment. This is the same logic that motivated EA's engagement with AI safety: the technology may well be coming regardless, and the responsible move is ensuring governance catches up.
On the argument for waiting. We have heard some arguments that it would be wiser to wait to invest in this space because it is possible that mitigation will fall more sharply, reducing the need to introduce new risks from SRM and similar technologies. We think these arguments are misguided for the same reason that the governance objection is. These technologies are feasible; it is better to invest now to ensure they are tested and governed than attempt to stop unwise deployments later. Further, emissions trajectories are plainly not sufficient to reduce risk below climate tipping point risk levels, meaning developing options now to address tipping points makes sense.
On arguments about AGI. We have heard some EAs argue that these risks are ancillary to risks of artificial general intelligence (AGI) or argue that AGI may somehow mitigate climate risks more directly. We are not AI or AGI specialists, but our take is as follows: First, climate overshoot risks greatly stress human economic and political systems. If one is focused on AGI risks, government capacity is important to manage them. It makes sense to develop options to protect against climate overshoot to preserve capacity for other issues – climate risk is a source of derailment risk on other threats. Second, the sorts of tools that could stabilize major earth systems are currently known; AGI, if present, may or may not accelerate research in this area, but scaling up research and governance from its current low base now still makes sense, as these tools remain core options for the climate risks to which we are committed. The universe of options is defined and will remain relevant. Investments at the scale we argue for will not pull resources substantially from other needs, but will create meaningful optional value on serious risks.
Summing up. Both SRM and ice stabilization have the potential to address major risks – and, if governed poorly – to cause major risks. They also receive very little attention, funding, and career dedication to manage these risks relative to their scale. But such attention could significantly expand government attention and facility with these issues. That is why we think investing now in proper governance, and research makes sense, and is highly tractable on the margin.
We think this analysis leads to a critical conclusion: The most pivotal intervention needed now is aligning governments to invest in climate stabilization research and governance, in time to unlock its option value. Funding that is marginal relative to the risk, on the order of tens of millions, would fundamentally alter the landscape here, and make risk reduction real.
As MacAskill observes, issues and political positions often have periods of "early plasticity, later rigidity." This is an uncommonly plastic time for climate stabilization. It is the right time to ensure that the many governments and civil society actors now facing major overshoot risks approach the issue in ways that are just, effective, and safe. That window is likely to close as political coalitions form policy positions in response to rising interest and overshoot risk. And several of those ways could lock in additional risk for humanity.
There are two distinct failure modes to worry about. First, it is possible that the window could close with most governments overlooking the option entirely because it has not been sufficiently raised into policy attention. That locks in the overshoot risks we have already discussed. Second, it is possible that, in the absence of good governance designs and research, some government during overshoot might react by attempting to deploy some of these technologies unilaterally, in the face of crisis. That could create increased risks of great power conflict, as could efforts by individual powers to accelerate their own capacities outside of stable government structures.
But we can head off these failures with effort, now, to construct align first mover governments around these issues. What is needed is both government capacity– the bureaucratic and fiscal ability of governments to act – and governance – a well-defined framework for risk reduction. Both are critical now, as the timing window closes. Here is our thinking:
First, deployment capacity sits with governments alone. Private actors can develop stabilization technology – at least one firm, Stardust, and several academic programmes, and now significant venture capital are doing so – but no private actor has legitimate authority to deploy at scale. Indeed, none could for any length of time without permission from the world’s governments, which have the legal and military capacity to stop private deployments and even private research. Decisions about whether, when, and how to use these tools are political decisions made by governments. A field with growing research capacity but no decision-making infrastructure is a field whose research cannot convert into risk reduction.
Second, the governance question is logically prior to the deployment question. Whether stabilization is safe to deploy depends on what governance is in place to evaluate the decision, constrain its scope, allocate liability, manage termination, and legitimate the outcome. Funders concerned about deployment risk – which includes most thoughtful EAs engaging with this space – should recognise that the risk profile of any future deployment is largely determined by the governance work that precedes it. There is no separate "safe deployment" lane that bypasses governance.
Third, governance is where current effort is smallest relative to the stakes. To be precise: we are not arguing research is crowded and governance empty. The wider ITN piece argues – and we agree – that the whole field is underfunded, including research. The narrower claim is that within the field, governance capacity is developing more slowly than research capacity, and the gap is widening as private capital moves into the technology side.
Why now. Political windows for building durable governance are closing, and it is worth being specific about what is closing them. Three dynamics to consider:
The combined picture is that governance is being shaped passively – by opposition and private capital – rather than actively by deliberative design. The window for change is narrow – and not just because of the climate bottleneck we are entering. Social and political windows are also closing (75% likelihood based on our assessment of recent spikes in political discussion, including efforts to ban or limit climate stabilization technology research in Africa, the U.S. and EU).The next 24-36 months determine whether a durable governance architecture gets built deliberately or improvised under crisis.
To close with a direct illustration of the option value advocacy to align governments on climate stabilization unlocks. Here is a structured sketch with assumptions readers can contest.
Probability of crossing at least one major tipping point (AMOC weakening or collapse, Amazon dieback, West Antarctic instability, or large-scale permafrost release) this century on current trajectories: we'd put this at 30-50% conditional on 2.5-3.5°C warming, drawing on recent multi-model assessments showing clustered tipping risk at these levels. Economic and welfare costs are genuinely difficult to bound – AMOC collapse alone would drop European temperatures and disrupt monsoons affecting billions – but plausible estimates run well into the tens of trillions in direct damages and far more in compounding effects on governance, food security, and displacement.
Within that broader picture, the Thwaites glacier illustrates where the stabilization argument is most concrete. Probability Thwaites contributes more than 0.5 metres of sea-level rise this century: ~40% given documented accelerating retreat. Direct damages conservatively $3-10 trillion, plus indirect costs from governance degradation and forced displacement. Expected cost of this component alone: $1.2-4 trillion. Glacier stabilization targeted at Thwaites specifically, if feasible, operates in the tens of billions. The risk-reduction ratio is large enough that even substantial uncertainty about feasibility does not close the expected-value gap.
We use Thwaites rather than AMOC for the concrete Fermi because it is the example where stabilization has an identifiable engineering intervention at a cost-able scale. AMOC is the more viscerally alarming example of the catastrophic-risk picture and lives in the Importance section above; Thwaites is the example where the per-dollar option-value calculation actually resolves. Both matter. Different parts of the argument rest on each.
The difference between stabilization deployed within robust governance versus in a crisis without it is large. Governance reduces the probability of termination shock, unequal regional distribution of effects, and great-power conflict over unilateral deployment. We would assign this a 10-30% reduction in expected harm across any future deployment scenario – across trillion-dollar stakes, itself a very large number.
Additional funding of roughly $100 million per year – the scale at which senior-level advocacy across key governments could materially shift governance frameworks – is a small fraction of the expected risk reductions above under almost any reasonable probability assignment.
Everything above grows further under EA-appropriate discount rates relative to standard government rates, and holds even at moderate rates given how much of the expected harm lands by mid-century rather than in 2100.
The shape of the calculation – modest costs against very large, tail-weighted stakes, with a small and responsive governance lever – is structurally similar to the calculation that motivated EA engagement with AI safety.
Here is where we are most and least confident, and what would shift our views.
The neglectedness case is the part of this argument we hold with most confidence. The funding numbers are verifiable and the comparison to AI safety funding in the early 2010s is not seriously contested.
The governance window claim we hold with high confidence (75% as we note above) but acknowledge is time–sensitive and could degrade quickly. If political positions have already locked in – if the governance window is effectively closed – the timing argument is less persuasive. We don't think this is true yet, but it is an empirical question.
The tractability of glacier stabilization is an open empirical question that a good feasibility study would substantially resolve. If the interventions prove technically infeasible at scale, the Thwaites case study becomes an argument for enhanced adaptation rather than stabilization.
The moral hazard evidence is favorable to our position, but the literature is genuinely mixed and large–scale political moral hazard – as distinct from individual–level – remains less studied. We think a sector–by–sector analysis, and region–by–region, analysis of the potential interactions between climate stabilization and emissions reductions in major economic sectors and global regions is critical, looking at both moral hazard and derailment risk. We think this is a plausible early investment opportunity.
The indirect longtermist pathway – climate overshoot degrading the governance substrate for AI safety and biosecurity – is the part of this argument we hold with the most uncertainty. It is qualitatively compelling but has not been formally modelled. The key crux: if the expected effect of climate overshoot on government capacity to manage existential risks is negligible even under pessimistic assumptions, this significantly weakens the importance case. Rethink Priorities is beginning work on this; it deserves serious engagement when published.
If you have expertise bearing on any of these cruxes – particularly on collapse threshold modelling, SRM governance tractability, or glacier stabilization engineering – we would particularly value pushback in the comments.
Following this logic, we have established a climate stabilization initiative, Climate Hub, composed of highly senior advocates from around the world, dedicated to raising the salience of these issues with governments to develop policy solutions. We do not pretend to have a monopoly on this issue – in fact, we think the best immediate path forward is to scale involvement across civil society and engaging existing institutions in facing these risks and possibilities squarely. We encourage others to explore these issues with similar urgency, as the climate precipice grows steeper. Approaches that would be productive include:
Scale up climate stabilization government capacity. Less than $15 million per year goes to the political and advocacy work required to build responsible climate stabilization research norms and governance frameworks. Senior–level advocacy operating across the US, India, China, and multilateral processes can mobilise significant government engagement for under $100 million annually. Governance decisions being made now about which research is permissible and which structures are legitimate will persist for decades. The literature suggests that “climate clubs” of early actor nations and civil society organizations can help shape governance durably – but no such club exists now on climate stabilization. One could, with sufficient advocacy.
Feasibility research for glacier stabilization. A rigorous engineering feasibility study for the full suite of proposed interventions doesn't currently exist. Funding one would cost in the range of tens of millions of dollars. This is a strong "giving to learn" opportunity.
Expanded research generally. The climate stabilization research community is very small relative to potential benefits, and there are many questions on the technology pathways for all these interventions that need work, and on critical sociological and economic questions, like how to make sure stabilization work supports long–term overall climate mitigation and carbon dioxide removal policies. Scaling up pathways for work here, including ensuring government budgets support long–term research communities that share knowledge, is critical.
Careers. The field has fewer than a few hundred researchers and advocates worldwide. EA-aligned people in climate science, political science, international relations, and governance have very high marginal impact here. This is a field where a small number of well-placed, analytically serious people have shaped the entire research and policy agenda.
Research on the crucial considerations. The collapse threshold question and the compound governance risk question are both potentially crucial considerations that could reshape climate stabilization's priority ranking. We encourage serious engagement with this work as it emerges.
Hey, welcome to the forum!
Just want to say I appreciate the deep dive and am going through it. My bias is that stabilization is worth funding and I already support Silverlining and now Reflective. I'd love to fund more of the space, but honestly it's been hard to find organizations ready to accept public donations and to come to weightings between SRM research and Glacier Stabilization (say).
(There was a great day-long event at SF Climate Week recently, btw - https://stabilize.earth/ ; were you there?)
Thanks for reading, and would love your take as you think this through -- drop me a line here or over linkedin/email. Seeing the crowd at SFCW was fascinating; really confirmed to me that this is an area that is set to move.
My sense (as you'd expect from the post) is that funding into the political/governance layer on top of the research development layer is a worthy additional path -- so both forwarding the great science work that Silverlining and Reflective (or Arete, say, for glaciers) do, and also looking towards orgs moving government capacity forward to shape policy that research informs. Looking forward to your review!