Security dilemma

Why can two rivals both spend more on defence and both end up less safe?

Suppose two neighbouring states, neither of which wants a war. Each looks at the other, decides it had better be able to defend itself, and spends accordingly. Ten years later both have spent a great deal, both face a stronger neighbour than before, and both feel less safe than when they started. Nobody lied, nobody planned aggression, nobody blundered. So what, exactly, did either of them do wrong — and if the answer is nothing, what kind of problem is this?

Start with a small question about hardware. Look at a tank, a fighter aircraft, a stock of artillery shells. What does the object tell you about the intentions of the state that bought it? Very nearly nothing. The same tank serves a plan to hold a border and a plan to cross one.

So an observer cannot read intent off capability. Could he read it off declarations? Only weakly. Intentions are invisible, cheap to state, and — this is the part that does the real work — revisable. Even a government that sincerely means what it says today cannot bind the government that replaces it. Capability, by contrast, is visible, and it persists.

Given that, what does prudence require of a defence planner? Plan against what the other side can do, not against what it says it will do. That is not paranoia; it is the only standard that can be checked. Now hold that thought and run both sides doing it at once.

A builds up, defensively. B, planning against capability, correctly observes that A's new forces could be used against him, and builds up in response — defensively. A observes B's build-up and finds his original fear confirmed. Each increment is honestly defensive in motive and genuinely threatening in effect, because effect is measured in capability and motive is not visible. Round it a few times: both have spent heavily, the relative balance may be roughly where it started, and both are now more frightened than before. That is the whole machine, and notice what it did not need. No aggressor, no deception, no stupidity.

That is why the word usually attached to it is tragedy rather than failure. John Herz, who named it in 1950, located the cause not in the actors but in their situation: with no authority above states to enforce a bargain, a promise not to attack has no enforcer, so nobody can safely rely on one. Moralising at the participants therefore changes nothing — each is behaving as a responsible government should.

But is it equally severe everywhere? Robert Jervis's contribution was to say no, and to identify two variables that set its intensity. The first is distinguishability: can preparations for defence be told apart from preparations for attack? If they can, a state that only wants to be safe has a way to prove it — it buys the distinguishable, defensive option and accepts that it now cannot attack. That is a costly signal, and it solves the observer's inference problem. The second is the offence-defence balance: does a given quantity of effort do more for the side attacking or the side holding? Where defence has the advantage, your neighbour's new brigade is much less alarming, because it takes a far larger effort to endanger you — and a smaller response restores your position.

Cross the two and you get four worlds, of which the worst is the one where postures look alike and offence pays. Then you cannot tell a defensive neighbour from an aggressive one, and waiting is dangerous, because whoever strikes first gains. That combination is often used to characterise Europe before 1914 — with the important caveat that what mattered was the widespread belief that offence paid, which turned out to be wrong.

Three refinements hold this together. The dilemma is driven by uncertainty about intentions combined with the dual-use nature of capability; remove either and it dissolves, which is exactly what transparency measures, verification regimes, and arms-control categories are trying to do. They do not appeal to goodwill. They try to make the observable facts carry information they did not carry before.

Distinguishability is a matter of degree, and of use rather than of the object: a fighter squadron reads as defensive until it is paired with aerial tankers and moved to forward bases, and the aircraft did not change. The offence-defence balance, though a useful idea, is hard to measure and has been criticised for being difficult to define independently of the outcomes it is meant to explain — so it is best used to compare situations, not to score them.

The last point is the one most worth keeping. That two states can arrive at hostility with no aggressor present does not establish that no aggressor is present. Some arms races are security dilemmas; some are one side arming to attack while the other responds. The framework explains how well-intentioned actors can end badly — it does not certify anyone's intentions, and used carelessly it becomes a way of explaining a real threat away.

• Why credible signalling requires an action that would be too expensive for a bluffer, and what that costs an honest state.
• How verification and inspection regimes attack the dilemma at its root by changing what is observable.

Every term the collection defines is gathered in the glossary.