An alliance exists to solve a problem of time inconsistency: "if you are attacked, I will come" costs almost nothing to say and a great deal to honour. Adversaries anticipate this, which is why a verbal guarantee produces no deterrence on its own.
Credibility can only come from sinking the cost in advance: forward-deployed troops (the "tripwire" — not there to win the fight, but to ensure any conflict automatically reaches the guarantor), integrated command, joint exercises. It amounts to hard-coding a key judgement into the process — and the cost is the same as with hard-coding: you have taken your own flexibility away.
| Arrangement | Structure | Cost / limit |
|---|---|---|
| Multilateral, symmetric NATO, 1949 Washington Treaty | Article 5 treats an attack on one as an attack on all | The text preserves discretion: each party takes "such action as it deems necessary," not automatic war. Invoked exactly once — 12 September 2001 |
| Bilateral, asymmetric US–Japan security treaty, signed and in force 1960 | The Article 5 defence obligation covers "the territories under the administration of Japan"; Article 6 provides bases | Not reciprocal defence but defence-for-bases — one side's autonomy and the other's depth of involvement are locked in together |
| Collective security The UN Charter system | Aimed at any disturber of the peace inside the system, not at a presumed external enemy | Activation requires great-power agreement, so it fails most readily when a great power is itself involved (see Day 11) |
Glenn Snyder set out the security dilemma internal to alliances in 1984: allies simultaneously fear two opposite things — abandonment (the partner does not come when it matters) and entrapment (being dragged into a war over the ally's quarrel). The two move inversely: reducing one necessarily raises the other.
The case for hardening commitments: vague promises get tested repeatedly, and deterrence fails precisely when an adversary misjudges resolve; saying it plainly and putting troops there is the cheapest defence against miscalculation.
The case for preserving flexibility: an automatic commitment outsources the decision to go to war to whichever member is most likely to get into a conflict. Christensen and Snyder described two symmetrical pathologies in 1990: chain-ganging (being dragged into war by a smaller partner) and buck-passing (everyone expects someone else to hold the line, so no one balances).
Assuming treaty text equals automatic entry into war. Most alliance treaties retain an interface along the lines of "in accordance with respective constitutional processes"; what determines credibility is the scale of the sunk costs, not the strength of the wording. Read the deployments, not the adjectives.
Deterrence works by having the other side calculate for itself that acting does not pay — so it is engineering aimed at an adversary's cognition, not at one's own strength. Snyder distinguished two routes in 1961: deterrence by punishment (the after-cost is more than you can bear) and deterrence by denial (what you want to achieve simply cannot be achieved). Three elements are all required: capability, credible resolve, and a signal that actually arrives.
The most counter-intuitive point: stability comes from ensuring your own retaliatory capability survives, not from being able to destroy the other side more thoroughly. When both sides' second-strike capabilities are reliable, striking first yields nothing, and the crisis pressure to "go first or lose out" disappears. Extended deterrence — pledging your costliest means on behalf of an ally — is discounted by its nature, which is exactly where fear of abandonment comes from.
The case for arms control: what it lowers is not only quantity but the probability of misjudgement — a standing verification channel is often the only communication line still working in a crisis.
The sceptical case: a treaty genuinely constrains only the party that intends to comply; in a multi-party environment a bilateral treaty may bind the compliant while leaving non-signatories free; and verification is never complete, so the residual uncertainty gets absorbed as a worst-case assumption.
Trade-off: signing buys predictability and sells the freedom to adjust unilaterally; not signing buys flexibility and sells the legibility of your own intentions — and legibility is the precondition for deterrence to work at all.
Equating deterrence with "more is safer": past assured second strike the marginal return falls away quickly, while the adversary's reaction is real. And equating arms control with disarmament — what it seeks is stability, which sometimes means restricting defensive systems, or even preserving weapons.
John Herz named the concept in 1950: state A strengthens its defences for self-protection; state B cannot tell self-protection from preparation for attack, so it strengthens too; A sees this and strengthens again. Both want only to be safe, and both end up less safe. The key point is that it requires no malice from anyone — it is a pure positive feedback loop, like two nodes each retransmitting on timeout and driving each other's congestion higher.
Robert Jervis identified two variables in 1978 that set how severe the dilemma gets: the offence–defence balance (whether attacking is cheaper than defending — where offence has the advantage, moving first pays) and distinguishability (whether intent can be read off equipment). Where both are favourable the dilemma nearly vanishes; where neither is, two sides slide into confrontation even with no expansionist aims on either part.
Jervis noted in 1976 that two models prescribe opposite responses to the same situation:
The spiral model: most confrontation grows from fear rather than greed. Showing strength confirms the other side's worst assumption and turns an avoidable conflict into a self-fulfilling prophecy; the prescription is verifiable signals of restraint.
The deterrence model: if the other side does hold expansionist aims, restraint and concession read as insufficient resolve and invite further probing — 1930s appeasement is the standard citation. The prescription is to draw clear lines early.
The hard part: which model applies depends on whether the other side is fearful or greedy — precisely the information that cannot be observed directly, and that changes over time. Misapplying the spiral model gets you eroded piece by piece; misapplying the deterrence model manufactures an adversary that did not previously exist.
Believing more communication dissolves it. The root is not a shortage of information but a commitment problem: even if everything the other side says today is true, no mechanism guarantees the next set of decision-makers will hold the same view, while the military capability remains. What tends to work is therefore not declarations but arrangements that are hard to reverse and externally verifiable — what they constrain is one's own future self.
Two frequently conflated concepts need separating first. Neutrality is a legal status in wartime: the 1907 Hague Convention V provides that a neutral state does not take part in the war and must not allow belligerents to move through or use its territory, while retaining the right to defend that status by force. Non-alignment is a peacetime political posture: not joining a military alliance, but carrying no reciprocal wartime obligations.
So "armed neutrality" is not a contradiction in terms: its credibility likewise comes from making violation cost more than it yields — the collateral is simply one's own defence rather than someone else's army.
| Route | How it was obtained | Cost / conditions |
|---|---|---|
| Recognition by the powers Switzerland, Congress of Vienna, 1815 | Permanent neutrality recognised by declaration of the assembled powers | Depends on external recognition continuing to hold; sustained by heavy militarisation and a militia system |
| Neutrality traded for sovereignty Austria, 1955 | The State Treaty restored sovereignty in May; the National Council passed the constitutional law on neutrality on 26 October | Neutrality was the price of the occupying forces' withdrawal — a negotiated outcome rather than a purely autonomous choice |
| Collective non-alignment The Non-Aligned Movement | The 1955 Bandung Conference (29 states) laid the groundwork; the first summit in Belgrade in September 1961 (25 states) founded it formally | Not equidistance between two blocs but a refusal to outsource sovereignty; collective scale raises bargaining power, at the cost of internal cohesion |
Status also changes: Finland and Sweden long practised military non-alignment and joined NATO in April 2023 and March 2024 respectively — evidence that this is a strategic response to the security environment rather than a cultural attribute.
The neutral / non-aligned case: it avoids being pulled into other people's quarrels (the entrapment risk from the first card) and preserves diplomatic room and a mediator's role; for a small state caught between great powers, joining one side may be what makes it the other's primary target.
The alliance case: neutrality is only as effective as others' willingness to respect it, and that is least reliable exactly when it is most needed — history is not short of neutrals that were marched through; and fully self-reliant defence is extremely expensive for a small state.
Trade-off: what each sells is exactly what the other buys — neutrality sells external guarantees in a crisis, alliance sells part of one's diplomatic autonomy. There is no universal answer, only one matched to geography and domestic consensus.
Reading neutrality as a moral stance, or as "offending nobody." What legal neutrality demands is strict formal parity — identical treatment of all belligerents — which at times cuts against moral intuition. A second misreading is that neutral states spend little on defence: those holding the status by their own strength typically carry a heavier burden.