A country announces a new missile-defence system. It is, its government says, purely defensive. The neighbour reads the same news and sees a shield that could let the first country strike without fearing the reply. So it orders more missiles. The first country sees those and orders more of its own interceptors. Nobody in this story needs to be aggressive. Everyone is trying to be safe, and everyone ends up less safe and poorer.
That pattern is so common that international relations (IR) scholars gave it a name, and built a discipline around the conditions that produce or soften it. This piece lays out the ideas a graduate keeps: why anarchy matters, what the main schools say in their own best terms, how deterrence works and fails, why wars happen even though they are costly, and where the democratic-peace argument stands today.
This describes how scholars explain relations between states. It does not tell you which country is right in any current conflict, and each school is given at its strongest.
Inside a country, if someone steals your car you can call the police, and the police answer to a government that holds a monopoly of force (see 35/13). Between countries there is no police force that can arrest a government. In IR the word for that is anarchy. It does not mean chaos. It means the absence of a ruler above the units. States still make treaties and often keep them; they just cannot be certain a treaty will be kept, and no one can be forced to keep it except by other states.
Kenneth Waltz's Theory of International Politics (1979) turned this into a full theory: leave aside what particular leaders or governments are like, and look only at the structure: anarchy, plus the way power is spread among states. Even without any villain, he argued, that structure pushes states to look after their own security. The older tradition, from Thucydides' account of the Peloponnesian War (fifth century BCE) through Hans Morgenthau's Politics Among Nations (1948), said much the same thing in different terms: the strong do what they can, and the weak accept what they must.
Other traditions noticed the same problem long before. The Arthashastra, a text of statecraft from India attributed to Kautilya (its dating is debated, but it belongs to the centuries around the turn of the era), describes a "circle of states" in which a neighbour is a natural enemy and a neighbour's neighbour a natural friend. The Chinese Warring States period (475 to 221 BCE) produced similar thinking. IR theory was largely written in Europe and North America after 1945, and scholars such as Amitav Acharya and Barry Buzan have argued since 2010 for a more global account that draws on these older sources too.
John Herz named it in 1950, and Herbert Butterfield described the same trap at about the same time: steps a state takes to make itself safer make others less safe, so they respond, and the first state ends up less safe than before. Robert Jervis's 1978 article "Cooperation Under the Security Dilemma" showed that how bad it gets depends on technology: if offensive and defensive weapons can be told apart, and defence has the advantage, states can protect themselves without threatening anyone. If offence has the advantage, whoever moves first wins, and every state is tempted to strike or to arm.
A game theorist reads that as a payoff table, and you can see the point on a small model. Two states each choose to Hold or Arm. If both hold, each gets 3. If both arm, each gets 1 (costly and no safer). If one arms and the other holds, the armer gets 2 + 1.5o and the holder gets 3 − 4.5o, where o is the offence advantage from 0 to 1. Slide o and watch which game appears.
Payoffs are illustrative, symmetric, and written as (row, column). Equilibria are computed by checking every cell for a profitable deviation.
Three worlds appear. When defence is strong (o below about 0.44), arming does not pay, and Hold is the best reply to everything: safe peace. In the middle band (from about 0.44 to two-thirds) there are two stable outcomes, mutual holding and mutual arming, and which one you get depends on whether each side trusts the other. This is the "stag hunt" that Rousseau described. Above two-thirds, arming is best whatever the other does, and both end up in the worse outcome: the prisoner's dilemma. That is the security dilemma at its sharpest. Historians still argue how much of the naval race between Britain and Germany before 1914 was this spiral and how much was German ambition (the Fischer debate), which is a reminder that the model describes one pattern, not every war.
Schools of IR are best understood as answers to one question: what drives state behaviour? Here is each in its strongest form.
These are not four teams in a league table. Scholars blend them, and a single event can be read through each. Pick one and compare.
Deterrence is persuading someone not to do something by threatening a cost that outweighs the gain. Thomas Schelling's The Strategy of Conflict (1960) and Arms and Influence (1966) made it a core part of the field. The point most people miss is that deterrence is not about how big the threat is. It is about whether the other side believes you will carry it out. A threat you would regret carrying out is a weak threat.
Put numbers on it. An attacker gains G if the attack goes unanswered, and pays a cost C if the defender retaliates. The attacker thinks the defender will retaliate with probability p. The attacker's expected value is (1 − p)G − pC. It is deterred when that is below zero, i.e. when p exceeds G ÷ (G + C). With G = 10 and C = 30, the defender only needs to be believed 25% of the time.
This is why bigger is not always better. A threat so terrible that carrying it out would also destroy the defender is a bluff everyone can see through, which is why states build things that make the threat credible: tripwire troops that guarantee a response, treaty commitments, second-strike forces that survive a first blow. Cold War strategists called the mutual version "mutual assured destruction". It is a standing argument among scholars how much nuclear deterrence in fact kept the peace between 1945 and 1991 and how much was luck: the Cuban missile crisis of October 1962 is cited by both sides of that argument.
War destroys wealth and lives. If both sides can foresee that, they should be able to agree on a deal that gives each what it would have won, minus the costs. James Fearon (1995) called this the puzzle: war should be avoidable, yet it happens. His answer: it takes one of a few conditions to break the bargain: private information and the incentive to bluff, commitment problems (a rising power cannot promise not to exploit its strength later), or issues that cannot be divided.
The model is tiny. Say the prize is worth 1. State A thinks it would win with probability pA; State B thinks A would win with probability qB. War costs A cA and B cB. A will accept a deal giving it at least pA − cA; B will offer up to qB + cB. A deal exists unless pA − qB is greater than the combined costs.
This also frames the old claim that trade prevents war. Norman Angell argued in The Great Illusion (1910) that war could not pay for modern industrial economies; war came in 1914 anyway. Liberals reply that trade raises the cost of war (the sliders' third preset) and defenders of the claim point to later evidence. Realists answer that interdependence also creates dependence that states try to escape or use as a weapon (Albert Hirschman, 1945), and that what matters is whether leaders expect a short, cheap war. Institutions are the second liberal answer: the UN (1945), NATO (1949), the EU's predecessors (1951 and 1957) and the WTO (1995) can share information and make commitments harder to break. Mearsheimer's 1994 article "The False Promise of International Institutions" replies that institutions mostly reflect the interests of powerful states. Both sides cite evidence, and the debate is not settled.
One of the most studied patterns in the field is that democracies hardly ever fight each other. The idea goes back to Kant (1795); the modern statistical case starts with Melvin Small and J. David Singer in 1976 and Michael Doyle's 1983 articles, and was strengthened by Maoz and Russett in 1993. Notice what it claims and does not claim: dyadic peace (between two democracies), not that democracies are peaceful in general. Democracies fight non-democracies, and have fought many wars of conquest and colonial wars.
Why might it be true? Two main explanations. Institutions: leaders who answer to voters pay a price for failure, so they pick fights more carefully and are more transparent. Norms: democracies expect each other to settle disputes by compromise. Scholars like Sebastian Rosato (2003) reject both, arguing the mechanisms do not work as claimed. Others say the pattern comes from something else: shared alliances during the Cold War, trade, or US dominance. A "territorial peace" argument reverses the arrow: settled borders produce both peace and democracy.
A statistical point decides how much weight to give the claim. The absence of war between two groups means little if there were few pairs to fight. Try it: of the 179 countries V-Dem counted for 2025, it classified 87 as democracies. How many country pairs are democracy-democracy? And if wars fell on pairs at random, how likely is it that none hit a democratic pair?
Simplified: wars land on pairs at random, which real wars do not (neighbours fight more). Democracy count for 2025 from the V-Dem Democracy Report 2026, as reported; other indices count differently, and long ago there were far fewer democracies.
At 87 democracies, about a quarter of all pairs are democratic, so 30 random wars would all miss them only about once in thousands of trials. With 5 democracies it would happen about 98% of the time. So the finding is informative only for the periods when democracies were numerous, and critics point out that the post-1945 sample is small and entangled with the Cold War. Supporters reply that controlling for contiguity, alliances, wealth and trade still leaves the effect. The status of the claim: widely accepted as a pattern, still disputed as a cause.
With no ruler above the states, safety has to be built by the states themselves, and the steps one takes can look like threats to another. Realists stress power, liberals trade, institutions and regimes, constructivists ideas and identity, and critics ask whose order this is. A threat deters only if it is believed, wars happen when bargaining breaks down, and the democratic peace is a real pattern whose cause is still argued.
Try this: next time a government says a military step is "purely defensive", write down how it looks from the other side of the border, then ask what each lens would say. If the two answers differ, you have found the security dilemma.
Sources. Waltz, Theory of International Politics (1979); Morgenthau, Politics Among Nations (1948); Herz (1950) and Butterfield (1951) on the security dilemma; Jervis, "Cooperation Under the Security Dilemma", World Politics 30(2), 1978; Mearsheimer, The Tragedy of Great Power Politics (Norton, 2001; updated edition 2014); Kant, Perpetual Peace (1795); Keohane, After Hegemony (1984); Wendt, "Anarchy Is What States Make of It", International Organization 46(2), 1992; Schelling, The Strategy of Conflict (1960); Fearon, "Rationalist Explanations for War", International Organization 49(3), 1995; Angell, The Great Illusion (1910); Hirschman, National Power and the Structure of Foreign Trade (1945); Mearsheimer, "The False Promise of International Institutions", International Security 19(3), 1994/95; Small and Singer (1976), Doyle (1983), Maoz and Russett (1993) and Rosato (2003) on the democratic peace; Acharya and Buzan, Non-Western International Relations Theory (2010). Democracy count: V-Dem Democracy Report 2026 (87 democracies out of 179 countries, 2025), as reported in coverage of the report. Course contents: MIT OCW 17.41 syllabus. Every number an interactive shows was computed in node: equilibrium bands at offence advantage 4/9 and 2/3; deterrence threshold G/(G+C) = 25% at G = 10, C = 30; bargaining ranges; C(87,2) ÷ C(179,2) = 23.5%; (1 − 0.235)30 = 0.03%.