Decreasing the Risk of Nuclear War Is Hard, but Possible

War is a fact of life. It is not a problem that can be fully solved with new inventions or math equations. Eliminating its prospect entirely may be impossible. While nuclear stalemate has made war between great powers less likely, the consequences of nuclear war would be devastating. Hiroshima and Nagasaki occurred when the U.S. was the only country with nukes—now there are nine countries with nuclear weapons, roughly 3,880 of which are deployed and ready to launch. 1 1 According to the Federation of American Scientists, the total inventory of nuclear weapons in 2024 is ~12,1109. Most of these weapons have explosive yields between 5 and 50 times more destructive than the weapons used on Hiroshima and Nagasaki, with some much smaller, and others up to hundreds and plausibly thousands of times larger. Expand Footnote Collapse Footnote More than once, it appears we have come right to the brink of starting a nuclear war. 2 2 Some of these close calls were more perilous than others, and it is often hard to know how close some of them were, given conflicting accounts, secrecy, and the fact that there has never been a two-sided nuclear war. One case with exceptionally high escalation risk was the Arkhipov incident during the peak of the Cuban missile crisis, where the crew of a Soviet submarine carrying a nuclear torpedo believed they were under attack. Facing temperatures from 50 to 65 C (122 to 149 F), high levels of CO2, U.S. signaling depth charges, blinding search lights, and numerous warning shots upon surfacing, there were multiple points where the submarine’s captain could have initiated nuclear use. Fortunately, despite the captain’s outbursts to prepare their nuclear torpedo, attack was either vetoed or talked down by Chief of Staff Vasilii Arkhipov. Expand Footnote Collapse Footnote Just because we have not yet experienced a nuclear bomb going off in our own country does not mean that the risk of a catastrophic war is low. In a world with nuclear weapons, the high absolute risks may be concealed by low probabilities of nuclear use. The difference between a 2% and 0.1% annual risk of nuclear war is imperceptible, but this difference in probability means the difference between expecting a nuclear war in your lifetime or avoiding it for centuries. 3 3 Note that while this logic holds true, compounding annual probabilities is likely not a good method for estimating an actual long-term probability of nuclear war, as the probabilities aren’t independent. This highlights the potential value of the “nuclear taboo”—nuclear war may become less likely the longer we avoid it. As of May 2024, prediction markets estimate a 20% chance of a non-test nuclear detonation by 2035. Meaning: an offensive nuclear detonation (intentional or inadvertent) or an accidental or unauthorized nuclear detonation. Expand Footnote Collapse Footnote

Risk is a function of probability and consequence. A 10% probability of losing $100 and a 50% probability of losing $20 have the same expected dollar value: a loss of $10. In the context of nuclear risk, far more significant consequences are at stake, but the principle remains the same. To decrease overall risk, we have to reduce the probability or severity of a bad outcome without making the other substantially worse. Roughly speaking: in order to improve our prospects, nuclear war must become less likely without becoming more deadly, or it must become less deadly without becoming more likely. 4 4 More precisely, lowering risk is lowering the resulting product of severity and probability. This formulation is based on expected value (probability x consequences = expected value). Expand Footnote Collapse Footnote

Lowering the risk of nuclear war is not straightforward. Attempts to decrease risk in one domain often transfer it to another as leaders adjust their strategies. This is called risk compensation: attempts to decrease risk are often offset by others’ reactions. Risk may remain constant, or perhaps gets worse, because changes to the risk landscape also change the incentives of various parties. This occurs because leaders don’t just care about the risk of nuclear war in a vacuum; they have other goals and will adapt to or manipulate risk to get more of what they want. Removing troops from an enemy’s border in the hopes of reducing the risk of war might just trigger an invasion.

Risk compensation is a common phenomenon that occurs across many domains. Everyday decisions like driving to the store involve accepting some risk of death, however small, to obtain a benefit. The initial introduction of football helmets may have increased lethal injuries as players were trained to make contact with the head due to the increased protection. Decreased personal risk resulted in a new playing style that increased the risk of playing football overall. When thinking strategically, it is essential to consider the ways in which new incentives might end up inadvertently rewarding more aggressive behavior or punishing the peaceful.

Unfortunately, many interventions which aim at only improving one factor, only the probability or consequences, can make total risk much worse. Imagine a magic switch which would either make the world's nuclear weapons one hundred times more powerful or a hundred times weaker. In the one case, nuclear weapons would be far more destructive to civilization, but less likely to be used given the consequences. In the other case, weaker nuclear weapons would still be a lot more powerful than conventional weapons: they might be more tempting to use in a crisis, making a devastating nuclear war more likely though less existential. Similarly, two common suggestions for reducing risk would likely encourage risk compensation: unilaterally increasing or decreasing the American nuclear arsenal. Imagine if the US destroyed all its nuclear weapons overnight: the destructive potential of a nuclear war might be lower, but the overall risk of global war would likely increase with such a one-sided reduction. On the other hand, if the U.S. doubled its arsenal in response to recent increases in China’s arsenal, it’s not obvious this would increase deterrence more than it would increase destructive potential, strategic instability, and further arms racing.

It is difficult to unilaterally lower risk because many efforts to do so provide incentives for other actors to engage in risk compensation. In the next section, we'll explore three general dynamics that make decreasing risk difficult.

The Difficulties of Decreasing Risk

Risk Unfolds Over Time

A crisis can end up decreasing total risk in the long run despite the massive increase in acute risk. For instance, the Cuban Missile Crisis was a scary moment in history—JFK estimated that we had a one in three chance of nuclear war. But imagine an alternative world where the U.S. didn’t react to Soviet deployments and each side maintained missiles in Turkey and Cuba. The acute risk from the crisis would have been removed, but so would some of the far-reaching beneficial outcomes. Would decades of having short time-of-flight missiles aimed at Washington and Moscow actually have been safer? Given this fragile status quo and an escalating arms race, any subsequent crisis could have been far riskier than the one we had. 5 5 Of course, it would have been better if the Soviet Union and U.S. had found a less threatening way to negotiate on missile reductions earlier and faster without any crisis occurring. Likewise, not all the lessons the U.S. and Russia took away from the crisis were necessarily adaptive: because the removal of U.S. missiles from Turkey was not public, many U.S. analysts likely overestimated the value of the U.S. confrontation. Likewise, the crisis may have pressed the Soviet Union to arms race faster. Expand Footnote Collapse Footnote

It is also important to hold bad actors accountable for their actions while allowing for the possibility of de-escalation. This is another reason why states should not always dodge potential crises—short-term passivity can encourage long-term risk taking. Off-ramps, which allow leaders to save face and avoid conflict, can incentivize future reckless military gambles by letting guilty parties off the hook. 6 6 Why is this the case? The discreet and indirect dialogues which can enable leaders to reach agreement often rely on plausible deniability and sometimes dishonesty toward outside parties. By preventing other parties from reaching common knowledge of what was agreed to, leaders can escape domestic and international pressure and work toward mutually beneficial deals. While this can be good when external political pressures are perverse, this poses multiple problems: 1) it is harder to coordinate with allies and other countries to punish defection on secret agreements and 2) it can allow leaders to make deals that aren’t in the interests of their populations. In general, increasing plausible deniability weakens norm enforcement, and allows salami slicing tactics and hybrid warfare to go unpunished. Expand Footnote Collapse Footnote When a crisis is not confronted head-on, leaders can start to feel as if there are no consequences to their actions. If the Cuban missile crisis didn’t occur, both sides might have felt like they could make larger provocations with impunity.

Reassurance and Deterrence

In the face of potential conflict, intentional deescalation can build goodwill and help avoid a downward spiral. But an aggressor may treat accommodation as a signal that one will take a beating without protest. For instance, in the early days of WWII, attempts to pacify and appease Nazi Germany failed to limit their expansionist intentions. Countries that engage in appeasement often seek to avoid a large-scale war, but can give their competitors the opportunity and confidence to expand piecemeal rather than face a united front from the start.

Too much reassurance without deterrence can be seen as weakness, but too much deterrence without any reassurance is just hostility. Efforts to deter an opponent which don’t credibly limit one’s own offensive intent can easily trigger security dilemmas that incentivize each side toward arms racing or pre-emptive attack. The worst combination possible is to have an absence of both short-term deterrence and long-term reassurance. Such a combination creates the perception of military opportunity now in order to prevent escalating hostilities and future doom. Whether it’s the way Putin seems to have perceived Ukraine, or the way the U.S. perceived Saddam’s hostility, WMD program, and military weakness in the years after 9/11, this combination leads to war. 7 7 In the case of Ukraine, Putin likely saw the West as divided and weak based on the Afghanistan withdrawal, and in its responses to Russia’s annexation of Crimea and military build-up near Ukraine. At the same time, fear of color revolutions, Western influence, and NATO enlargement made the status quo less desirable. Putin’s perceived opportunity to grow his legacy, distract from domestic political problems also likely contributed to his motivations to invade. In the case of Iraq, its weak performance in the first Gulf War presented an opportunity for U.S. military victory while Saddam’s selective transparency in ending his WMD program and his initial response in the days following 9/11 conveyed extreme hostility and grave long-term risk. Expand Footnote Collapse Footnote

Stability–Instability Paradox

The stability-instability paradox is the observation that when both sides of a conflict are well-equipped with nuclear weapons the probability of a nuclear war between them decreases, but the probability of smaller confrontations, skirmishes, and proxy conflict increases. 8 8 See Glenn Snyder’s The Balance of Power and the Balance of Terror and Robert Rauchhaus’ Evaluating the Nuclear Peace Hypothesis. Expand Footnote Collapse Footnote Mutual fear of nuclear war causes both sides to avoid major provocations, while giving them confidence that minor provocations will not escalate. Indo-Pak skirmishes & terror attacks, the Ukraine war, and Chinese opposition to certain risk reduction measures all reflect this logic. It isn't always good to just increase one dimension of strategic stability: countries must think of how competition will be redirected to determine if total risk has been reduced. As the US, Russia, and China continue modernizing their nuclear arsenals, how secure each feels about their nuclear forces will have an impact on the kinds of conflicts they are willing to fight. The greater their comfort at the nuclear level, the greater instability at lower levels of conflict.

Michael Cummings, Daily Express, August 24 1953, "Back to Where it all Started"

What Can We Do?

While nuclear risk reduction can be quite complex it isn’t hopeless. It is possible for countries to pursue strategies that are more robustly ethical and less susceptible to risk compensation by other states. 9 9 Ethical in the sense of reducing expected loss of life. In other words, such ideal solutions would be a moral Pareto improvement. Expand Footnote Collapse Footnote

Risk compensation is easiest to avoid where there is no incentive or opportunity to engage in it. If the other side doesn’t know risk has been reduced, they won’t have an incentive to act more aggressively. De-escalation plans undertaken in secret are much harder for an opponent to exploit than ones articulated in public. 10 10 If, in complete secrecy, countries took steps to prevent extreme escalation internally, then the worst outcomes could be prevented. Even if countries knew that each other were likely to have escalation control plans, there would be little they could do about it if the plans remained secret. If successful, secret escalation controls reap the benefits of nuclear stalemate without the dire consequences if something were to go wrong. Expand Footnote Collapse Footnote Leveraging AI as a “fog of war machine” to optimize the concealment and survivability of nuclear forces, countries may be able to stably reduce their number of nuclear weapons. 11 11 On the other hand, if a country fears that the other side may outsmart their algorithm, they might still face pressure to arms race. Expand Footnote Collapse Footnote Even without secrecy or deception, unilateral risk reduction interventions can target gaps and biases in the other side’s intelligence gathering efforts, thereby reducing risk in their blindspots. 12 12 In a sense, the other side’s ignorance or distraction is a substitute for secrecy. Just as military strategists can leverage the other side’s cultural, ideological, or bureaucratic blindspots for a military advantage at a stable level of risk, peace strategists can leverage the same vulnerabilities to lower risk while sustaining the status quo. Expand Footnote Collapse Footnote

Another path to reducing risk without emboldening the other side to more aggressive behavior, is to concentrate on who risk is being reduced for. Ideally, keep the risk high for key decision makers, or what they care about, while limiting harms to civilians. Such bespoke deterrence strategies that target what dictators care about, while aiming to mitigate harm to millions of civilians, may help decrease the probability and severity of war. 13 13 A related but more general concept is the idea of “tailored deterrence” which aims to customize deterrence strategies for particular leaders, groups, and contexts. Expand Footnote Collapse Footnote Advances in sensing, stealth, hypersonics, and precision may enable such strategies. 14 14 Mobile and hardened targets were traditionally hard to destroy without nuclear weapons due to the required combination of speed, large blast yield, and large blast area. If AI and sensing reduce uncertainty about exact target locations, while hypersonic and penetrating weapons provide greater capability, far fewer targets would still require a nuclear response (or, at least, a high-yield one). However, pursuing such weaponry without providing reassurance against surprise attack or pre-emptive use could greatly increase the risk of false alarm and inadvertent escalation. This could produce “use it or lose it” incentives in crises where such weapons put the other side’s nuclear arsenal, leadership, or command and control systems at risk. This highlights the problem of “entanglement” where a country mixes its nuclear and non-nuclear forces in a manner that makes it harder to tell if an incoming attack is limited or incapacitating in nature. Expand Footnote Collapse Footnote Rather than a nuclear weapon being dropped on a city, precise conventional weapons can target specific buildings, weapons, or vehicles. By still threatening key targets, such strategies could reduce reliance on gratuitous and imprecise nuclear weapons without weakening deterrence. At the same time, similar advances could lead to “use it or lose it” incentives in crisis and further arms racing as leaders come to fear the prospect of new kinds of disarming attacks. 15 15 An alternative proposal that aims to address this concern is the idea of limiting nuclear targeting to conventional military forces and war-supporting industry rather than civilian targets, leadership, or nuclear forces. The concept attempts to strike a complex balance: targeting something dictators care about (maintaining their power), while minimizing harm to innocents, arms races, and the “use it or lose it” incentives posed by directly targeting leadership or their nuclear weapons. Nevertheless, the strategy isn’t without trade-offs. Targeting war supporting industry would still massively affect civilians, and should limited nuclear war begin, holding to the policy would allow leaders to continue the conflict without direct threat to themselves. For more on the complexities, see here. Expand Footnote Collapse Footnote

Finally, there can sometimes be opportunities to incentivize cooperation while disincentivizing defection. 16 16 Within game theory, this involves changing the payoff matrix to make cooperation more attractive and defection less. Expand Footnote Collapse Footnote As previously mentioned, it's challenging to maintain a balance between providing reassurance and showing strength—achieving this equilibrium is tough but feasible. In the late Cold War, the Reagan administration simultaneously pursued mutual disarmament with the Soviet Union, while demonstrating the ability to continue arms racing if necessary. At that point in time, the Soviet Union knew that it was incapable of beating the U.S. in an arms race and began mutual disarmament. 17 17 Given their economic disadvantage and recent loss of industrial spies, Soviet near-term prospects for continued competition were dim, and this also forced Gorbachev’s efforts at broader reform. But there were still rough patches in Reagan’s strategy. Before the Soviets developed a better understanding of Reagan’s intentions, his initial “Evil Empire” framing of the Soviet Union contributed to their existential fears that the U.S. was preparing for a surprise attack, which may have greatly heightened the risk of accidental nuclear war earlier in his administration. See Hoffman’s book The Dead Hand for more. Expand Footnote Collapse Footnote One must simultaneously demonstrate a credible commitment to a mutually beneficial peace, and a conditional threat if such peace is not secured.


Risk compensation is similar to the conservation of energy: risk has a tendency to be conserved across situations. Attempts to reduce risk often feel hopelessly complex since they involve balancing multiple interacting factors simultaneously. But careful analysis might reveal opportunities to reduce risk without sacrificing one’s own strategic position. Strategies which avoid risk compensation are meant as a supplement to traditional diplomacy. Even in the presence of antagonism or conflict, solutions to risk compensation are often still available. In an ideal world, either side would be able to implement these strategies without relinquishing their strategic position. Such a gain would be purely moral—all else being equal, more lives saved is a good thing.


Article co-written by Matthew Southey.

About the author

Matthew Gentzel

Matthew co-leads Longview Philanthropy's nuclear weapons policy program and co-manages Longview’s Nuclear Weapons Policy Fund. Before Longview, Matt researched emerging tech policy and threat assessment at OpenAI and the National Security Commission on AI. He recently published Countering Competitive Risk Compensation: Principles for Reducing Nuclear Risk on Net.