Deterrence in International Relations

What Deterrence Is

Deterrence is the strategy of preventing an adversary from acting by convincing him, in advance, that acting will cost more than it is worth. Everything decisive happens before the event, and none of it happens in the physical world. The missiles sit in their silos; what moves is the adversary’s estimate of consequences. This is why the decisive variable is never what you have but what the other side believes you will do with it.

• The word comes from the Latin de terrere, to frighten away — and the etymology is the mechanism, not decoration.
• Deterrence dissuades an actor from an action not yet begun. Success looks like nothing happening at all.
• It threatens a cost the adversary himself concludes is not worth paying, rather than physically preventing him.
• It is relational, not possessional. A state deters this adversary from this action; a posture that works on one may leave another unmoved.
• It assumes rationality — ordered preferences, cost-benefit calculation, a preference for not losing. Every serious criticism attacks this assumption somewhere.

Deterrence, Compellence, Defence and Détente

Thomas Schelling drew the distinction that organises the field: deterrence persuades an adversary not to start; compellence persuades him to stop, reverse, or begin something he does not want to do. Deterrence asks for inaction, which is invisible and open-ended. Compellence asks for visible movement by a deadline, and visible movement under threat is indistinguishable from public submission.

• Compellence is harder, for a structural reason. Compliance must be seen, so it carries a reputational price the target may rate above the punishment threatened.
• Deterrence has no deadline; compellence must have one, or there is nothing to comply by and the threat decays into noise.

Glenn Snyder made the second distinction canonical. Defence denies the adversary his objective if he attacks, operating after the decision to attack and judged by whether the attack succeeds. Deterrence prevents the decision by raising anticipated cost, and is judged by whether an attack is ever launched. A state can be strong in one and weak in the other.

Response	Logic	Operates	Succeeds when
Deterrence	Raise the anticipated cost	Before the adversary decides	Nothing happens
Compellence	Impose cost until he reverses	After action has begun	He visibly stops or undoes
Defence	Deny him the objective	After the attack begins	The attack fails
Détente	Reduce the hostility itself	Continuously, politically	He stops wanting to act

• Détente is the third traditional response, working on the political relationship so that the intention to attack never forms.
• The 1970s détente produced the architecture deterrence then rested on — SALT I, the ABM Treaty, the Helsinki Final Act — so accommodation and deterrence are complements, not rivals.

The Three Requirements

Deterrence is usually stated as capability plus credibility plus communication and then left as a list. The list is only useful unpacked, because each element fails differently and each is repaired by different means. None is guaranteed by possessing weapons.

Capability

• Capability means the ability to inflict damage the adversary regards as unacceptable — a threshold set by the target’s values, not the threatener’s arsenal.
• “Unacceptable damage” is a political judgement. Robert McNamara‘s figure of a quarter to a third of Soviet population and half its industry was an administrative convenience for sizing forces, never a discovered constant.
• Sufficiency, not superiority, is the criterion, since beyond destroying an adversary once, extra warheads add nothing to the threat.
• Survivability is what matters. A force destroyable in the adversary’s first strike is not a deterrent but a target, and its vulnerability tempts attack.
• It is bought by hardening, dispersal, mobility, concealment and multiple platforms.
• Capability without survivability inverts deterrence into provocation — Albert Wohlstetter‘s central insight.

Credibility — the Hardest of the Three

Credibility is the belief that the threat will actually be executed. It is hardest because nuclear retaliation, at the moment it would be carried out, is irrational: the victim of a first strike gains nothing by destroying the attacker’s cities and invites destruction of what remains of its own. A threat everyone knows would be irrational to execute is not obviously believable, and this is the commitment problem at the heart of the theory.

• Resolve is the perceived willingness to bear costs, inferred from past behaviour, declared interests and domestic politics rather than from capabilities.
• Reputation is treated as a stock that can be spent and rebuilt — why states fight over intrinsically worthless ground, on the argument that backing down anywhere discounts every threat everywhere.
• Credibility falls as the stake falls. Threatening retaliation for an attack on one’s own capital is believable; for a distant ally’s border dispute it is not.
• Threatening something manifestly irrational is not credible, so the theory’s real work is making an irrational threat believable anyway.
• Schelling’s answers to the commitment problem are the field’s most influential body of thought, and each works by reducing the threatener’s own freedom of choice.
  • The threat that leaves something to chance. Rather than promising to launch, a state creates a situation carrying a genuine uncontrollable risk of catastrophe and threatens to raise it. Nobody must believe you would choose annihilation — only that things might slip.
  • Tying hands. A commitment becomes credible by being costly to break: a tripwire garrison whose destruction forces a response, a pledge embedded in a treaty and a public declaration.
  • Burning bridges. Removing one’s own retreat, so that fighting is the only remaining course — the general who destroys the crossing behind his army.
  • The rationality of irrationality. It can be rational to cultivate unpredictability, because a leader who might do something insane cannot safely be pushed; the pose’s danger is that one’s own system believes it.

“The power to hurt is bargaining power. To exploit it is diplomacy — vicious diplomacy, but diplomacy.” — Thomas Schelling

• The stability-instability consequence follows directly. If the nuclear threat is credible only for the largest provocations, everything below that level is effectively unthreatened and the adversary is free to operate there.

Communication

• A misread threat does not deter, however capable and however credible the threatener believes it to be.
• Declaratory doctrine is the primary instrument — published doctrines, posture reviews and no-first-use pledges exist to tell adversaries what triggers what.
• Exercises, tests and deployments say what words cannot, demonstrating that the capability is real and the organisation can use it.
• Ambiguity cuts both ways. Deliberate vagueness about thresholds widens the range an adversary must fear and preserves flexibility; clarity reduces the risk that he miscalculates where the line is.
• Israel’s undeclared arsenal is the purest ambiguity posture; India’s published no-first-use doctrine is near the opposite pole.

Deterrence is the only military capability whose entire output is an idea held by someone else, which is why its hardest problems are problems of belief rather than of engineering.

Types of Deterrence

Punishment and Denial

• Deterrence by punishment threatens retaliation the adversary cannot prevent — take what you want, and lose something you value more. Nuclear retaliation against cities is the archetype.
• Deterrence by denial threatens to make the objective unattainable — try, and fail. Air defences, hardened infrastructure, forward-deployed forces and cyber resilience work this way.
• The worked contrast: the American promise to retaliate for a Soviet invasion of Western Europe was punishment; the NATO armoured divisions that would have to be fought through were denial.
• Denial is more credible and less escalatory, since executing it is simply defence and needs no decision to destroy a population.
• Punishment is cheaper and scales better, which is why states with limited resources and vulnerable geography reach for it.
• Denial has grown with precision weapons, and now anchors thinking about deterring territorial seizure in the Indo-Pacific and eastern Europe.

General and Immediate

• Patrick Morgan separated deterrence as a standing condition from deterrence as crisis management.
• General deterrence is continuous and unfocused — a standing threat maintained against any future attack, operating with no crisis in view. It is what an arsenal does every day.
• Immediate deterrence operates in a crisis, when attack is imminent or in preparation, aimed at a specific action in a specific window.
• The worked contrast: the Cold War as a whole was general deterrence; the thirteen days of October 1962 were immediate deterrence.
• General deterrence failing produces a crisis; immediate deterrence failing produces a war.

Direct and Extended

• Direct deterrence protects the deterrer’s own territory, where the interest at stake is survival and credibility is therefore highest.
• Extended deterrence protects a third party, and is what turns an arsenal into an alliance asset.
• It is intrinsically less credible. The threatener is asked to accept destruction of its own cities to avenge a foreign one, and the adversary knows it.
• The Cold War formulation: would Washington trade New York for Bonn? — a question no American administration ever answered to European satisfaction.
• NATO’s answers were institutional rather than logical: forward-based American troops as a tripwire, nuclear sharing placing American weapons on allied soil under dual-key control with allied delivery aircraft, and a Nuclear Planning Group giving allies a voice in targeting.
• Japan and South Korea live entirely under extended deterrence, having forgone arsenals in exchange; South Korea now records open public majorities for indigenous weapons.
• The current European debate is the sharpest test since the 1960s. Doubts about American reliability produced, on 2 March 2026, President Macron’s framework of “advanced deterrence” — an expansion of the French arsenal, temporary deployment of nuclear-capable Rafale aircraft to allied bases for exercises, and a structured strategic dialogue with European partners.
• France retains sole command throughout and has stopped publishing warhead numbers to restore strategic uncertainty, so what is offered is participation rather than a guarantee.
• A French umbrella faces that discount more harshly, its arsenal being smaller and its doctrine historically tied to French vital interests alone. Alliances as such belong to the companion article Balance of Power.

Central, Theatre, Minimum, Maximum, Existential

• Central deterrence concerns homeland-to-homeland exchange using intercontinental systems; theatre deterrence covers a defined region with shorter-range systems, and exists because central threats are not credible for regional stakes.
• Minimum deterrence holds that a small survivable force able to destroy a few of the adversary’s cities suffices, since no rational leader risks even that. It accepts that nuclear war can be prevented but not fought.
• Maximum, or war-fighting, deterrence holds that credibility requires the ability to fight and prevail at every level, demanding large, accurate, counterforce-capable arsenals.
• Recessed or non-weaponised deterrence is capability demonstrated but weapons unassembled and undeployed — India’s own position from 1974 to 1998.
• Existential deterrence, associated with McGeorge Bundy, argues that the mere existence of the weapons deters regardless of doctrine or posture, because no leader can be sure a crisis will not end in their use. If true, most doctrinal apparatus is theatre.

Cross-Domain, Integrated and Conventional

• Cross-domain deterrence threatens retaliation in a different domain from the attack — sanctions for a cyber intrusion, a conventional strike for an attack on satellites.
• It exists because domains are asymmetric, so the threat must be issued where the threatener is strong rather than where the attack occurred.
• Cyber deterrence is the hardest case. Attribution is slow and contestable and effects are hard to calibrate, so denial through resilience has largely displaced punishment in this domain.
• Space is becoming a domain of its own, since satellites carry the early warning, command and navigation on which every other deterrent depends — making them simultaneously the most tempting and the most escalatory targets.
• Economic deterrence — sanctions, export controls, financial exclusion — inflicts punishment without fighting, and its use against Russia since 2022 is the largest experiment yet in whether it deters anything.
• Integrated deterrence is the American formulation of the 2020s: coordinating nuclear, conventional, cyber, space, economic and allied instruments into a single cost calculus rather than separate ladders.
• Conventional deterrence predates nuclear weapons and outlives them. John Mearsheimer‘s work argues it fails precisely when an attacker believes a quick decisive victory is available, so force posture matters more than force totals.
• The nuclear revolution sat on top of conventional deterrence — and the stability-instability paradox is what happens at the seam.

Distinction	First term	Second term	The discriminating question
Mechanism	Punishment — retaliate after	Denial — defeat the attempt	Impossible, or merely costly?
Time	General — standing	Immediate — crisis-specific	Is an attack already in prospect?
Beneficiary	Direct — own territory	Extended — an ally	Whose survival is at stake?
Scope	Central — homeland exchange	Theatre — regional	Intercontinental or regional systems?
Size	Minimum — enough to punish once	Maximum — enough to prevail	Preventable only, or winnable?
Domain	Single-domain	Cross-domain / integrated	Must the answer come from where the attack came?

The Intellectual History

Deterrence is an ancient practice — walls, garrisons and reprisal threats are as old as organised violence — that acquired a formal theory only when a weapon appeared whose use could not be survived. The theory was built almost entirely between 1946 and 1966, largely by civilians rather than soldiers, in response to a technical fact: the bomb made defence impossible and therefore made prevention the only remaining strategy.

Brodie and the Founding Formulation

• Bernard Brodie stated the nuclear revolution in a single sentence written within months of Hiroshima, reversing the purpose of armed force.

“Thus far the chief purpose of our military establishment has been to win wars. From now on its chief purpose must be to avert them. It can have almost no other useful purpose.” — Bernard Brodie

• The reversal is not rhetorical. If war between nuclear powers destroys the winner, victory ceases to be a coherent war aim.
• His operational formulation is that a credible nuclear deterrent must be always ready but never used — permanently at a readiness that makes the threat believable, permanently unfired.

Schelling and Strategy as Bargaining

• Schelling redefined strategy itself. It could no longer be the science of military victory; it had become the art of coercion, intimidation and bargaining, in which force is valued for the influence it buys rather than the ground it takes.
• The “diplomacy of violence” is his term for the capacity to hurt used as a currency of negotiation, deployed before and instead of fighting.

Kahn and the Escalation Ladder

• Herman Kahn insisted on thinking about the unthinkable, on the argument that refusing to analyse nuclear war did not make it less likely, only less controllable once begun.
• His escalation ladder ran from political crisis through conventional war, demonstration detonations and limited exchange to spasm war, treating escalation as a series of discrete choices rather than a single fall.
• The ladder’s purpose was deterrent, not permissive: intermediate rungs make intermediate threats credible, so a state is not forced to choose between nothing and everything.
• The objection it attracted was moral — that treating megadeaths as a variable normalises them — and it stands against the whole genre.

Wohlstetter and the Delicate Balance of Terror

• Wohlstetter’s argument: the balance of terror is not automatic. Mutual possession does not produce stability; stability requires that both sides’ forces survive a first strike.
• His basing studies showed American bombers were vulnerable, which meant the deterrent could in principle be disarmed — and a disarmable deterrent invites the attack it exists to prevent.
• The policy consequences were enormous: hardened silos, airborne alert, dispersal, and above all the ballistic-missile submarine, the only leg never reliably locatable.

Snyder, George and Smoke, Morgan

• Glenn Snyder separated deterrence from defence as distinct logics with different force requirements, and formulated the stability-instability paradox.
• Alexander George and Richard Smoke studied deterrence failures rather than successes, and found that deterrence typically fails not because the threat was too weak but because the challenger acted around it — probing at levels the threat did not cover, or believing the commitment did not extend to his case.
• Patrick Morgan built the general and immediate distinction into the standard apparatus and insisted deterrence be studied as a relationship sustained over time.

The Game-Theoretic Apparatus

• The standard games are the field’s shorthand for its recurring structures.
  • The prisoner’s dilemma models the arms race: arming is each side’s dominant strategy whatever the other does, and the rational outcome — mutual armament at great cost — is worse for both than the restraint neither can trust.
  • Chicken models crisis confrontation, two players approaching collision with the winner taking the prize and both destroyed if neither yields; it illuminates October 1962 and every standoff since.
  • The stag hunt models arms control: joint restraint is best for both, but fear of defection makes each settle for the safe inferior option.
• Game theory as a method is developed in the companion article Systems Theory and Scientific Approaches to International Relations and is not re-taught here. What matters is only that rational actors, each behaving sensibly, reliably produce collectively irrational outcomes.

Doctrinal Evolution Across the Cold War

Nuclear doctrine did not evolve because strategists grew wiser. It evolved because each doctrine, once stated, revealed a credibility gap the adversary could exploit, and the next was built to close it.

• Massive retaliation, announced by Secretary of State John Foster Dulles in 1954, promised response instantly, by means and at places of American choosing — deliberately unspecified, so any aggression anywhere risked general nuclear war.
• Its defect was credibility, and the critique was decisive: it offered a president only capitulation or holocaust, and nobody believed a superpower would choose holocaust over a minor territorial adjustment. Soviet intercontinental delivery capability then destroyed it entirely.
• Flexible response, adopted under Kennedy and developed by McNamara, supplied graduated conventional, tactical, limited and general options so that a proportionate response always existed; NATO adopted it formally in 1967.
• Its own defect was ambiguity, and European allies feared it meant a conventional war fought on their territory rather than escalation to the American homeland.
• Assured destruction was McNamara’s force-sizing criterion: retain, after absorbing a first strike, the capacity to inflict unacceptable damage in return.
• Mutual assured destruction is what results when both sides achieve it — a condition rather than a policy, never formally adopted by anyone, and the frame within which arms control was negotiated.
• Countervalue targeting aims at what the adversary values — cities, population, industry — and needs relatively few, relatively inaccurate weapons, which is what makes minimum deterrence workable.
• Counterforce targeting aims at his military forces, above all his nuclear forces, and needs many warheads, high accuracy and good intelligence.
• The strategic asymmetry reverses the moral one. Countervalue is the more monstrous threat and the more stabilising posture; counterforce is the more discriminating threat and the more destabilising posture, because a force built to destroy an arsenal looks exactly like a force built to strike first.
• Counterforce creates use-it-or-lose-it pressure on the side that fears disarmament, which is how accuracy became a threat to stability.
• The Schlesinger Doctrine of 1974 introduced limited nuclear options — small, selective, largely counterforce strikes — so that a president had choices between nothing and destroying Soviet society; critics replied that making nuclear use thinkable makes it likelier and that no exchange could be shown to stay limited.
• The countervailing strategy of Presidential Directive 59, 1980, extended this by targeting what the Soviet leadership valued most — political control, command bunkers, military power — so that no Soviet theory of victory at any level could succeed.
• The Strategic Defense Initiative, announced in 1983, attacked the foundation rather than the doctrine, proposing a shield to render nuclear weapons “impotent and obsolete”.
• SDI was destabilising, exactly as MAD’s logic predicts. An imperfect shield is useless against a full first strike but useful against a degraded retaliation — valuable, that is, only to a state intending to strike first, and Moscow read it that way.

Doctrine	From	Core claim	Gap closed	Gap opened
Massive retaliation	1954	Any aggression risks general nuclear war	Cheap deterrence of a larger army	Not credible for small provocations
Flexible response	1961	Graduated options at every level	Credibility at low levels	Allied fear of a conventional war at home
Assured destruction / MAD	1960s	Neither side escapes retaliation	First-strike incentive	Nothing deters sub-nuclear conflict
Limited nuclear options	1974	Selective counterforce available	Presidential choice below general war	Makes nuclear use thinkable
Countervailing strategy	1980	Deny victory at every level	Soviet theories of limited victory	Requires very large accurate forces
Strategic defence	1983	Defend rather than retaliate	Escape from mutual vulnerability	Destroys the other side’s second strike

Mutual Assured Destruction in Full

MAD is nuclear balance, and equally the balance of terror. Its content is one proposition: each side convinces the other that starting a war is starting a suicide. Nobody chose it; both arrived at it, and once arrived at, a rational first strike became impossible — which made it simultaneously stabilising and morally intolerable.

Second Strike, Survivability and the Triad

• Second-strike capability is the ability to absorb a first strike and still inflict unacceptable damage, and it is the technical requirement without which MAD does not exist.
• The nuclear triad is delivery by three independent platforms — land-based intercontinental ballistic missiles, submarine-launched ballistic missiles, and air-delivered weapons.
• Three legs, three vulnerabilities, so no single technological breakthrough or surprise attack eliminates all three.
  • Land-based ICBMs are accurate, responsive and under continuous positive control, but their locations are known and fixed — the most targetable leg, and therefore the leg that absorbs an attacker’s warheads.
  • Submarines are the survivable leg, a ballistic-missile boat on patrol having never been reliably located; they are the true guarantor of the second strike, at the cost of slower communication with national command.
  • Bombers are recallable and visible, and therefore the signalling instrument — launched as a demonstration of resolve and turned around, which no missile can be.
• Not every nuclear state maintains a triad: Britain has reduced to a single sea-based leg and France maintains sea and air legs, while Russia, the United States, China and India maintain all three.

Why Defence Destabilises

• The central paradox: defending yourself makes everyone less safe. If one side can intercept the other’s retaliation, the second strike is no longer assured and MAD dissolves.
• A defence is most useful against a ragged retaliation — precisely what remains after striking first — so possessing one signals an intention its possessor may not hold.
• The rational response to a defence is more offence: more warheads, decoys, multiple independently targetable re-entry vehicles, and manoeuvring or hypersonic delivery.
• The Anti-Ballistic Missile Treaty of 1972 therefore limited defences deliberately, restricting each party to two deployment areas and later one, and banning development and testing of sea-, air-, space- and mobile land-based systems.
• The clearest case of adversaries agreeing to stay vulnerable because vulnerability was safer — and its abandonment in 2002 removed the keystone of the whole architecture.

Launch on Warning and the Accident Risk

• Launch on warning means firing while the adversary’s missiles are still in flight, on early-warning sensor data, before any warhead detonates.
• Launch under attack means firing after the first detonations are confirmed but before one’s own force is destroyed — marginally later, still measured in minutes.
• Both exist to protect vulnerable land-based missiles, and both compress national decision-making to roughly ten to fifteen minutes.
• They turn a false sensor reading into a possible war, and are simultaneously rational responses to vulnerability and the largest single source of accidental-war risk in the system.
• SIPRI’s January 2026 figures put 2,100–2,200 warheads on high operational alert, almost all Russian and American, which is the size of the standing risk.

The Near-Misses as Evidence

The near-misses are usually told as anecdotes. They are better read as data, because each tests a different link in the deterrence chain and each shows a link that nearly gave way.

• The Cuban missile crisis, October 1962, tested crisis stability with immediate deterrence running in both directions, and showed how far events escape leaders’ control — a Soviet submarine officer’s refusal to concur in a nuclear torpedo launch, unknown to either capital, being the thinnest margin the era recorded.
• Able Archer, November 1983, tested communication and failed it. A NATO command-post exercise simulating nuclear release was read by a Soviet leadership already primed for surprise attack as possible cover for a real one, and Soviet forces were alerted during what the West believed was a routine drill.
• The Petrov incident, September 1983, tested launch on warning. A Soviet early-warning satellite reported an incoming American launch; the duty officer, Stanislav Petrov, judged it a malfunction and did not pass it up. It was sunlight on high cloud.
• The pattern is the same: deterrence held, but through individual judgement operating outside the system rather than through the system — a weaker finding than the record of non-use suggests.
• The record is nonetheless remarkable, no nuclear weapon having been used against an adversary since August 1945, and the near-misses are the reason it cannot confidently be credited to deterrence alone.

The Stability-Instability Paradox

Snyder’s paradox is the most useful idea available for reading contemporary nuclear rivalries. If nuclear weapons make major war between two states effectively impossible, both know that whatever they do to each other will not escalate to nuclear exchange — and that knowledge makes limited conflict safer, and therefore more attractive, than it was before.

• The mechanism is a transfer of risk, not its abolition. Stability at the highest level of violence purchases instability at lower levels.
• The nuclear ceiling shelters proxy war, sub-conventional conflict, terrorism and limited operations with a confidence that would be reckless between non-nuclear rivals.
• The Cold War evidence is substantial: no direct superpower war, and a continuous series of proxy conflicts from Korea to Afghanistan.

South Asia as the Working Case

India and Pakistan are the standard illustration because the sequence is unusually clean: an unresolved territorial dispute, a contiguous border, four wars, and declared arsenals since 1998. What has happened since is what the paradox predicts.

• Kargil, May–July 1999, came within a year of the tests. Pakistani regulars and irregulars occupied heights across the Line of Control; India mounted a large conventional operation but refused to cross the Line of Control or the international border — restraint without precedent in 1965 or 1971.
• The 2001–02 standoff followed the attack on Parliament in December 2001. India mobilised in Operation Parakram, roughly half a million troops faced each other for ten months, and the forces withdrew without a war and without a political result.
• The Cold Start debate arose from that failure — the Army’s search for rapid, shallow, multiple-thrust offensives launchable in days rather than the three weeks Parakram took, to find usable conventional force below Pakistan’s nuclear threshold.
• Pakistan’s answer was to lower the threshold. Short-range, low-yield tactical nuclear weapons, exemplified by the Nasr system, are designed to make any Indian incursion, however shallow, run into nuclear use — closing the space Cold Start sought to open.
• Uri, September 2016: after an attack on an army camp India announced surgical strikes across the Line of Control. Publicising them was the innovation; similar operations had been conducted unacknowledged before.
• Balakot, February 2019: after Pulwama, India struck from the air into Pakistani territory proper for the first time since 1971, followed by an aerial engagement and the return of a captured pilot. Another threshold was crossed and the conflict again ended within days.
• Operation Sindoor, May 2025, is the furthest either side has gone.
  • The trigger was the Pahalgam terror attack of 22 April 2025, which killed 26 people; India responded by placing the Indus Waters Treaty in abeyance.
  • In the early hours of 7 May 2025 India struck nine sites in Pakistan and Pakistan-occupied Kashmir associated with Lashkar-e-Taiba, Jaish-e-Mohammed and Hizbul Mujahideen.
• The exchange ran roughly four days. Pakistani drone and missile attacks on Indian cities and installations were met by strikes on Pakistani air defences and, on the morning of 10 May, on air bases including Nur Khan at Rawalpindi, Murid, Rafiqui and Sargodha.
• It ended on 10 May 2025 in an understanding reached between the two Directors General of Military Operations.
• India’s declared conclusion was doctrinal. In his address of 12 May 2025 the Prime Minister set out a “new normal” in three parts.
  • Decisive retaliation for any terrorist attack, striking terror infrastructure at its source.
  • No distinction between terrorists and the states that shelter them.
  • A refusal to be deterred by nuclear blackmail — an explicit statement that Pakistan’s arsenal no longer buys immunity for sub-conventional attack.
• The counter-argument: this is the belief the paradox makes dangerous. Confidence that the ceiling holds is what permits each successive escalation, and nobody has established where the ceiling is.

Four days of conventional exchange between two nuclear-armed states is not a failure of deterrence. It is deterrence working exactly where the paradox said it would, and nowhere else.

Waltz and Sagan

• Kenneth Waltz argued that nuclear weapons kept the Cold War cold: they make the costs of war certain and obvious, leaders therefore become cautious, and caution is what peace requires.
• His stronger claim was that selective proliferation may stabilise, new nuclear states facing the same discipline and learning the same restraint — “more may be better”.
• Scott Sagan replied from organisation theory, and the reply is not about rationality but institutions: nuclear weapons are managed by large military organisations with parochial interests, standard routines, secrecy and imperfect learning.
• Such organisations produce accidents routinely, and the American record of false alarms, lost weapons and unauthorised handling — accumulated by the best-resourced system in the world — is the evidence.
• Military organisations also carry biases against stability: a preference for offensive doctrines, for pre-delegated launch authority, and for secrecy that defeats civilian oversight.
• New nuclear states are worse placed — thinner civilian control, weaker safety cultures, shorter distances, less warning time, contested internal authority.
• The South Asian record supports both readings. Waltz can cite a quarter-century of nuclear peace and visible restraint at every crisis; Sagan can cite Kargil, a warning time measured in minutes, tactical weapons requiring pre-delegated release, and the shortness of the record.
• The sample is too small. Eighty years of non-use is a suggestive record, not a demonstrated law, and the events that would falsify Waltz need happen only once.

Arms Races and the Domestic Sources of Armament

Deterrence requires arms, and arms acquisition has a political economy of its own. The standard account treats arms racing as a response to external threat; the more uncomfortable account is that much of it is generated inside the arming state, by constituencies whose interests are served whether or not the threat is real.

• Eisenhower’s farewell address of January 1961 described a permanent armaments industry of a size new in American experience.

“In the councils of government, we must guard against the acquisition of unwarranted influence, whether sought or unsought, by the military-industrial complex. The potential for the disastrous rise of misplaced power exists and will persist.” — Dwight Eisenhower

• His point was structural, not conspiratorial. No cabal was required; a large permanent defence sector, absent in peacetime before 1945, would generate its own political weight.

Model	Driver	Prediction	Weakness
Action-reaction	The adversary’s build-up	Symmetrical, responsive spending	Cannot explain build-ups when threats fall
Domestic structure	Bureaucratic, industrial and legislative interests	Spending persists regardless of threat	Understates real external danger
Technological imperative	What laboratories make possible	Qualitative races; capability seeking a mission	Treats politics as passive

• The action-reaction model is the classical one, formalised in Lewis Fry Richardson‘s equations, where each side’s armament rate rises with the other’s stock and falls with its own economic burden. It is the arms-race face of the security dilemma, developed in the companion article Security.
• The domestic-structure model reads spending through bureaucratic politics, where a service’s budget share is what is actually being fought over, and organisational process, where procurement follows existing routines rather than fresh threat assessment.
• The technological-imperative model observes that weapons are often built because they can be and a mission is found afterwards — multiple warheads, missile defence and hypersonic systems all began as laboratory possibilities.
• Defence contractors hold concentrated, organised interests in continued procurement against a diffuse public interest in restraint — the classic asymmetry of collective action.
• Legislative constituencies matter because production is deliberately distributed across many electoral districts, so cancellation is a local employment decision in dozens of places at once.
• Inter-service rivalry generates requirements: each service’s claim on the nuclear mission is simultaneously a claim on budget share, and the triad is partly a bureaucratic settlement.
• Worst-case threat assessment is institutionally rational for the assessor, who bears a career cost for underestimating an adversary and none for overestimating one, and it systematically inflates the threat.
• Revolving-door employment between services, defence ministries and contractors makes the three sets of interests hard to separate in practice.
• The consequence for arms control is direct. Negotiations must overcome not only the adversary’s mistrust but the domestic coalition profiting from the status quo, which is why agreements are so often bought with new programmes.
• The current evidence is unambiguous. World military expenditure reached $2,887 billion in 2025, up 2.9% in real terms and 2.5% of world GDP.
  • The top five were the United States at $954bn, China $336bn, Russia $190bn, Germany $114bn and India $92.1bn, the top three alone accounting for 51% of the global total.
  • European spending rose 14% to $864bn, the sharpest growth since the Cold War ended.
• All nine armed states are modernising simultaneously, an outcome no action-reaction model alone accounts for.
• Declared policy has tracked the hardware. The United States Nuclear Posture Review of February 2018 reaffirmed continued reliance on nuclear weapons for its own security and that of its allies and partners: it added supplemental low-yield options — a low-yield submarine-launched ballistic-missile warhead and a nuclear sea-launched cruise missile — and extended the “extreme circumstances” in which use would be considered to cover significant non-nuclear strategic attacks.
  • The 2022 review softened the edges without changing the premise. It cancelled the sea-launched cruise missile and retired the B83 bomb, but rejected both a sole-purpose and a no-first-use declaration and retained extended deterrence for allies as a core mission.
  • No later review has replaced them. In March 2026 the American defence department stated that it did not intend to conduct a formal Nuclear Posture Review, treating the 2018 document as the operative statement of policy — so continued reliance on nuclear weapons, for its own and its allies’ security, remains the declared position of the state with the largest deployed arsenal.

Arms Control and Its Erosion

Arms control manages the competition rather than ending it. Its aim is not disarmament but stability: reducing the incentive to strike first, the probability of accident and the cost of the rivalry, while leaving deterrence in place.

Instrument	Year	What it did
Partial Test Ban Treaty	1963	Banned tests in the atmosphere, outer space and under water; underground testing continued
Non-Proliferation Treaty	1968	Froze the club at five; non-proliferation, peaceful use and disarmament as three pillars
SALT I and the ABM Treaty	1972	Capped strategic launchers; deliberately limited defences to preserve mutual vulnerability
SALT II	1979	Equal aggregate ceilings; signed but never ratified after Afghanistan, though observed
INF Treaty	1987	Eliminated an entire class of ground-launched missiles of 500–5,500 km range
START I	1991	First deep cuts in deployed strategic warheads, with intrusive verification
START II	1993	Would have banned multiple-warhead land missiles; never entered into force
CTBT	1996	Bans all nuclear explosive testing; has never entered into force
SORT	2002	Cut deployed warheads to 1,700–2,200 with almost no verification machinery
New START	2010	1,550 deployed strategic warheads and 700 delivery vehicles each, with on-site inspection

• The erosion began in 2002, when the United States withdrew from the ABM Treaty to pursue missile defence — the first abandonment of a major strategic treaty, and the removal of the provision that made mutual vulnerability deliberate.
• The INF Treaty collapsed in 2019, the United States citing a prohibited Russian cruise missile and Russia citing American launchers in Europe; an abolished class of weapons became legal again.
• The Open Skies Treaty, permitting unarmed observation flights over members’ territory as a confidence-building measure, lost the United States in 2020 and Russia in 2021.
• Russia de-ratified the CTBT in November 2023, matching the American position of signature without ratification and leaving the global testing moratorium resting on political undertakings alone.
• In October 2025 the American president ordered a resumption of nuclear weapons “testing”, officials later suggesting subcritical experiments were meant, without White House confirmation; the United States then cast the sole vote against a General Assembly resolution supporting the moratorium. No explosive test has occurred, the last American one being in 1992.
• New START expired on 5 February 2026 with no successor. For the first time since 1972 there is no negotiated limit on the size or composition of the two largest arsenals, and no inspection regime covering them.
• What followed is reassurance without obligation.
  • Russia announced it would observe the treaty’s central limits only so long as the United States did not exceed them.
  • The United States said it sought a “modernised” treaty covering all Russian warheads and Chinese growth; China declined to join, arguing the two largest holders must reduce first.
  • Informal understandings were discussed and none was concluded.
• SIPRI’s assessment is that the vacuum, alongside universal modernisation, undermines the legitimacy of the NPT itself, because the disarmament pillar has visibly stopped moving.

For the first time in more than half a century the two largest arsenals are limited by nothing but the intentions of the governments that own them.

Disarmament and a Nuclear-Weapons-Free World

• The Non-Proliferation Treaty rests on three pillars — non-proliferation by states without weapons, a right to peaceful nuclear technology, and disarmament by states with them.
• Article VI binds every party to pursue negotiations in good faith on ending the arms race and on nuclear disarmament. It is an obligation of conduct that has never been enforced.
• The bargain is discriminatory by design, permanently dividing the world into five recognised weapon states — those that tested before 1 January 1967 — and everyone else.
• India’s objection is to the discrimination, not to disarmament. A treaty legitimising indefinite possession by five while prohibiting it to all others is neither equitable nor a disarmament instrument, and India has not signed.
• Indefinite extension in 1995 converted a twenty-five-year arrangement into a permanent one, sharpening that objection considerably.
• The Eleventh Review Conference closed on 22 May 2026 without a consensus outcome — sixteen years since the last agreed document, with disarmament benchmarks, testing, transparency, references to the prohibition treaty and Iran the principal fault lines.
• The International Court of Justice’s advisory opinion of 1996 held that the threat or use of nuclear weapons would generally be contrary to the law applicable in armed conflict, with one case it could not resolve.

“…the Court cannot conclude definitively whether the threat or use of nuclear weapons would be lawful or unlawful in an extreme circumstance of self-defence, in which the very survival of a State would be at stake.” — International Court of Justice

• The Court was unanimous on a second point usually overlooked: an obligation exists to pursue in good faith and bring to a conclusion negotiations leading to nuclear disarmament in all its aspects.
• The Treaty on the Prohibition of Nuclear Weapons entered into force on 22 January 2021, banning development, possession, stationing, use and threat of use outright, with 74 states parties and 25 further signatories as of late 2025.
• What it achieves is normative; what it does not achieve is material. No armed state and no state under an extended-deterrence umbrella has joined, so not one warhead is covered; its function is to stigmatise possession and deepen the humanitarian framing.
• The CTBT’s entry-into-force problem is structural. It requires ratification by 44 named states holding nuclear technology; the United States, China, Egypt, Iran and Israel have not ratified, and India, Pakistan and North Korea have not signed.
• The export-control regimes are the non-treaty half of the architecture — the Nuclear Suppliers Group, the Missile Technology Control Regime, the Wassenaar Arrangement and the Australia Group. India belongs to the last three; Chinese insistence on a criteria-based route open to Pakistan too has kept it out of the NSG.
• Global Zero argues for phased, verified, multilateral elimination on a fixed timetable, on the ground that non-proliferation cannot survive indefinitely alongside permanent possession.
• The decisive intervention in this argument came in January 2007, when George Shultz, Henry Kissinger, William Perry and Sam Nunn — four Cold War policymakers with impeccable deterrence credentials — published a newspaper article reversing their position and calling nuclear weapons a source of extreme risk rather than of safety.
• Their reasoning concerned the new context: many more possessors, weaker safeguards, shorter warning times, states without the accumulated safety practice of the superpowers, and terrorist groups openly seeking a device.
• Rajiv Gandhi’s Action Plan, presented to the UN General Assembly in June 1988, is India’s own most substantial disarmament initiative — a phased programme binding on all states for the elimination of all nuclear weapons by 2010, with a commitment not to develop new ones and a demand that the regime’s discrimination end.
• It tied disarmament to equity. India was not asking to join the club but to dissolve it on a schedule, and when that was refused it eventually went overt.

Prospects, Honestly Assessed

• Four obstacles stand between the world and zero.
  • Verification is the first. Confirming that no warheads remain, in states with decades of production records and large fissile stockpiles, demands an intrusiveness no great power has accepted.
  • Breakout is the second. At zero, the first state to reconstitute even a handful of weapons holds a monopoly, so the incentive to cheat rises as numbers fall and the last stage of disarmament is the most dangerous.
  • Security guarantees are the third. States hold weapons because they fear something, so abolition requires resolving the disputes that generate the fear rather than legislating around them.
  • Conventional imbalance is the fourth and least discussed. Removing nuclear weapons restores the advantage of whoever has the largest conventional forces, which is why states in weak conventional positions value them most.
• The direction of travel is the wrong way. Every armed state is modernising, China’s arsenal is growing, the strategic treaty regime has expired, the testing norm is under open pressure, and the review process cannot agree a document.
• The counterweight: non-use has held for eighty years, and possessors have multiplied far more slowly than mid-century forecasts predicted. Abolition is not close; drift toward use is not established either.

The Critiques of Deterrence Theory

• The rationality assumption is the first target. Deterrence needs an adversary who calculates, values survival and controls his own forces; leaders who are ideologically committed, badly informed or hostage to domestic politics may fail all three tests.
• Kissinger’s objection is the crispest version: deterrence applies only to rational actors, so where rationality cannot be assumed, states must find other means of reducing the threat.
• Robert Jervis supplied the psychological critique from inside the field — leaders systematically misperceive, reading others’ defensive measures as offensive, fitting ambiguous evidence to existing beliefs, and overestimating both their own clarity and the adversary’s coherence.
• Success is unobservable. One cannot show why something did not happen; a peaceful year is equally consistent with successful deterrence, with an adversary who never intended to attack, and with restraint produced by something else.
• The moral objection is that deterrence holds civilian populations hostage as settled policy, forming an intention to kill people with no influence over their government — an intention just-war reasoning cannot license even if never executed.
• Nina Tannenwald’s nuclear taboo is the constructivist alternative explanation for non-use. What has restrained states since 1945, on her account, is not the cost calculus but a normative prohibition that made nuclear use illegitimate rather than merely imprudent.
• Her evidence is where deterrence cannot explain restraint: the United States did not use nuclear weapons in Korea or Vietnam against adversaries who could not retaliate, and the reason was the anticipated political and moral cost of breaking the tradition of non-use.
• The implication for non-nuclear states is that strengthening the norm may serve them better than acquiring weapons.
• Russian nuclear signalling since 2022 has made the taboo’s strength a live dispute rather than a settled finding.
  • Repeated invocations of nuclear readiness over Ukraine, exercises with non-strategic systems and a revised doctrine lowering the declared threshold show that the threat of use has been rehabilitated as an instrument.
  • Whether that erodes the taboo, or demonstrates it because the threats were made and not executed, is genuinely contested.
• Deterrence fails against non-state actors. A network with no territory, no population and no fixed assets offers nothing to hold at risk, its leadership may welcome martyrdom, and attribution may take months.
• What remains is denial rather than punishment — hardening, intelligence, interdiction, resilience — plus deterring the states that shelter them, which is exactly the logic of India’s “new normal”.
• Proliferation risk is the theory’s own by-product. If nuclear weapons deliver security, every insecure state has a reason to acquire them, and each acquisition gives its neighbours the same reason.

Pre-emptive and Preventive War

The two terms are routinely used interchangeably and mean quite different things. The distinction decides whether an act is arguably lawful self-defence or straightforwardly aggression.

• Pre-emption responds to an attack that is imminent — prepared, in motion, about to land — and strikes first only in the sense of firing before the blow arrives.
• Prevention attacks a capability before it matures, on a judgement that the adversary will become dangerous later. There is no imminent attack, only a projected one.
• The legal positions differ accordingly. The Caroline formula requires a necessity of self-defence instant, overwhelming, leaving no choice of means and no moment for deliberation; Article 51 of the UN Charter preserves self-defence “if an armed attack occurs”. Pre-emption strains this framework; preventive war has no accepted basis in it at all.
• The Bush doctrine of 2002 collapsed the distinction deliberately, arguing that against adversaries with weapons of mass destruction and no return address, waiting for imminence was waiting too long — and using the language of pre-emption for what was preventive logic.
• The 2003 Iraq war discredited it. The projected capability did not exist, demonstrating at enormous cost that preventive war rests entirely on intelligence about a future that has not happened.
• The Osirak precedent of 1981, Israel’s strike on an Iraqi reactor, is the case cited in favour and is genuinely ambiguous — condemned by the Security Council at the time, now widely credited with delaying an Iraqi programme.
• The Iranian case is the live one.
  • Israeli strikes on Iranian nuclear facilities began 13 June 2025, and American strikes on Fordow, Natanz and Esfahan followed on 21–22 June 2025.
  • The European three triggered snapback on 28 August 2025, UN sanctions were reimposed at the end of September, and further strikes came in February–March 2026.
  • As of March 2026 Iran still held roughly 200 kg of 60%-enriched uranium.
• The case shows prevention’s limit. Facilities can be destroyed; knowledge, personnel and dispersed stockpiles cannot, and the political effect of being attacked strengthens rather than weakens the argument for acquiring a deterrent.
• The strongest argument is technical. A disarming strike must destroy every warhead, and a mature triad — mobile launchers, submarines at sea — cannot be comprehensively located, let alone destroyed at once.
• Prevention works only before a survivable arsenal exists, which makes it a counter-proliferation tool rather than an alternative to deterrence, and one that accelerates the proliferation it targets.

Deterrence Now

• The bipolar frame has gone, and the mathematics with it. The theory was built for a two-player game; the United States, Russia and China form a three-body problem in which every measure against one adversary alters the balance with the other.
• China’s expansion is the driver. Its arsenal, around 620 warheads in early 2026, is growing faster than any other and could exceed American or Russian ICBM numbers by 2030.
• The global picture is about 12,187 warheads, some 9,745 in military stockpiles and about 4,012 deployed, with Russia and the United States holding roughly 83% of stockpiled weapons.
• Missile defence and hypersonic delivery reopen the ABM Treaty’s question without its answer: defences invite offensive expansion, and manoeuvring hypersonic systems are designed to defeat defences and compress warning further.
• Artificial intelligence in nuclear command and control is the newest risk, because the premium on speed pushes decision support toward automation exactly where human judgement has historically been the last safeguard.
• Cyber threats attack the nervous system, not the weapons, and corrupted early-warning or command networks undermine confidence in the retaliation deterrence depends on.
• Entanglement of conventional and nuclear systems is a distinct danger: shared early-warning satellites, dual-capable missiles and co-located command facilities mean a conventional attack may be indistinguishable from the opening of a nuclear one.

The Indian Anchor

• India’s posture from 1974 to 1998 is the textbook case of recessed, non-weaponised deterrence — demonstrated capability, unassembled weapons, deliberate ambiguity about whether any existed.
• The tests of May 1998 ended the ambiguity, framed as a response both to the security environment and to the indefinite extension of a regime offering India no path to recognition.
• The draft doctrine of August 1999, from the National Security Advisory Board, set out credible minimum deterrence, no first use, a triad and civilian control, and served as the public statement of intent for four years.
• The Cabinet Committee on Security adopted the operative doctrine in January 2003, and it remains formally unchanged.

The 2003 Doctrine

• Four components define the declared posture.
  • Credible minimum deterrence — an arsenal sized to inflict unacceptable damage and no larger, with credibility taking priority over numbers.
  • No first use — nuclear weapons to be used only in retaliation for a nuclear attack on Indian territory or on Indian forces anywhere.
  • Massive retaliation — a first strike to be answered with retaliation inflicting damage the attacker cannot accept, which forecloses limited nuclear exchange by declaration.
  • The exception for chemical or biological attack — India retains the nuclear option against a major attack with those weapons, the one qualification on no first use written into the doctrine itself.
• “Minimum” is a relational judgement, not a number. It means enough surviving warheads to inflict unacceptable damage after absorbing a first strike, so the figure rises as adversaries’ counterforce capabilities improve — which is why a minimum posture is compatible with continuous growth.
• No first use costs India something real. It requires absorbing a first strike before responding, and therefore far greater survivability, dispersal, hardening and secure command; it forgoes any disarming option; and it makes the arsenal useless against conventional threats.
• What it buys is crisis stability and political standing — less pressure to launch on warning, lower risk of accidental use, firmly civilian command, and a sustainable claim to be a responsible power with a genuine disarmament record.

Forces and Command

• The triad is now substantially complete. INS Arihant was commissioned in August 2016 and made its first deterrence patrol in November 2018; INS Arighaat followed in August 2024 and INS Aridhaman on 3 April 2026.
• The K-series submarine-launched missiles run from the K-15 at 750 km, deployed, through the K-4 at 3,500 km, tested but not confirmed deployed, to the K-5 at 5,000 km in development — range being the constraint on genuine second-strike depth.
• The land leg is anchored by the canisterised Agni-V, over 5,000 km, MIRV-tested in March 2024, with Agni-Prime in service; canisterisation shortens launch times and raises readiness.
• India’s stockpile is estimated at about 190 warheads, against Pakistan’s 170 and China’s 620.
• The Nuclear Command Authority comprises a Political Council chaired by the Prime Minister, which alone can authorise use, and an Executive Council chaired by the National Security Adviser; the Strategic Forces Command manages the arsenal.

Is No First Use Eroding?

• Rajnath Singh’s remark of August 2019 — that India had adhered strictly to no first use but that what happens in future depends on circumstances — is the standing official hint of a rethink. It changed nothing on paper and was never withdrawn.
• Christopher Clary and Vipin Narang argue that the capabilities India is acquiring point toward a counterforce option: accurate missiles at higher readiness, canisterised and possibly pre-mated systems, greatly improved satellite and drone surveillance, and layered missile defences to catch a ragged Pakistani response.
• The destabilising consequence is the point. If Pakistan believes India is acquiring a disarming option, it faces use-it-or-lose-it pressure in any crisis and both sides gain incentives to move first — the exact condition second-strike survivability exists to remove.
• The counter-case: the evidence is circumstantial, the same capabilities being explicable as survivability and readiness improvements under no first use, and a disarming strike on a dispersed Pakistani arsenal remaining technically implausible.
• India’s treaty position is consistent across all three instruments: outside the NPT as discriminatory, outside the CTBT absent a time-bound disarmament framework though maintaining a voluntary testing moratorium, and outside the TPNW as an instrument negotiated away from the disarmament machinery.

Pakistan’s Answer

• Pakistan’s posture is full-spectrum deterrence, an explicit rejection of both no first use and minimum deterrence, covering the range from battlefield use to strategic retaliation.
• Its logic is conventional inferiority. Unable to match Indian conventional forces, Pakistan threatens early nuclear use to deny India any usable conventional option — the reasoning NATO applied against Soviet numbers in Europe.
• Tactical weapons are the instrument, and they carry the costs the category always carries: forward deployment, pre-delegated release authority, and dispersal in a crisis, which weakens central control precisely when control matters most.
• Each side’s solution is the other’s problem — India’s search for conventional space produces Pakistan’s tactical weapons, which produce India’s counterforce temptation, which produces Pakistan’s dispersal and expansion.

The whole architecture of nuclear peace now rests on beliefs that no treaty verifies and no institution supervises.