Skip to content
NatureHQ

Ecologyecological relationship

Predator–prey arms races

Nobody is racing. Each side changes because the other changed — and where the prey can kill you, the usual asymmetry disappears.

Neither side is trying to do anything. Each changes because the other has changed, over generations, and the outcome of any given encounter stays uncertain. The stakes are usually uneven: a prey animal that fails dies, a predator that fails goes hungry.

The metaphor is useful and dangerous in equal measure, and the paper that popularised it said so at the time. Useful, because reciprocal selection is real: a change in prey defence alters what works for the predator, which alters what works for the prey, and the process can run for a long time. Dangerous, because "arms race" imports intention, planning and an escalation towards some conclusion, and none of those belong. No newt is manufacturing toxin to defeat snakes. Newts vary, the more toxic ones survive encounters more often in places where snakes eat newts, and toxicity rises in those populations across generations. The snakes are doing the mirror image. Nobody is racing. Two things make this more than a definitional caution. The first is the asymmetry: selection on prey to escape is generally stronger than selection on predators to catch, because failure costs the two sides different amounts — the life–dinner principle. That predicts prey should often be ahead on any trait where both are investing, which is broadly what is observed. The second is the exception, which is where the evidence is best. Where prey are genuinely dangerous, a failed attack can kill the predator too, and the asymmetry weakens. Rough-skinned newts carry tetrodotoxin; some garter snake populations have evolved resistance to it; and crucially the two vary together geographically, with the most toxic newts occurring where the most resistant snakes are. That geographic matching is what turns a plausible story into evidence, because a one-sided explanation does not predict it.

Early coverage · 36% complete · reviewed 2026-09-03

What this page covers

Reciprocal escalation has been documented in a small number of systems studied in unusual depth — garter snakes and newts above all — and inferred much more widely. The inference is often reasonable and is not the same as the demonstration.

Often confused with: A predator and prey both being good at something, which may reflect no reciprocal selection at all; Directed evolution, as though either lineage were aiming at the other; A race with a finish line — escalation stops when costs bite, and frequently reverses

Quick facts

What it actually means
Reciprocal selection across generations, with no goal on either side
The life–dinner principle
Prey that fails dies; a predator that fails misses a meal
The best-evidenced case
Newt toxicity and garter snake resistance vary together across populations
Not endless
Escalation stops where the cost of more exceeds the benefit

What the metaphor gets wrong

The paper that popularised the term also warned about it.

Prey do not evolve defences in order to defeat a predator. Each side changes because the other side has changed, over generations, with nothing aiming at anything.

Established

Specialists would state this without hedging. Multiple independent lines of evidence agree.

Predator–prey arms races describe reciprocal selection: change in one party alters the selective environment of the other, producing escalation without foresight or goal-directedness in either lineage. The process is typically asymmetric, since failure costs prey their lives and predators a meal.

Who this applies to
A general evolutionary process, best documented in a small number of intensively studied systems.
Studied in
Animalia
Why we rate it this way, and what the caveats are
EstablishedHigh confidence

The evolutionary logic is standard and the reciprocal pattern has been demonstrated with matched geographic variation in at least one system, which is what distinguishes an arms race from a story about one.

How far it can be extended

Reciprocal escalation has been documented in independent systems, most thoroughly in the garter snake and newt pairing.

Caveats

  • The asymmetry has exceptions: where prey are dangerous, a failed attack can kill the predator, and the stakes even out.
  • Escalation is not inevitable or endless. Costs, constraints and the arrival of other predators all interrupt it.

Still unanswered

  • How often apparent arms races are genuinely reciprocal rather than one lineage responding to a stable feature of the other.

Last reviewed 2026-09-03

The evidence (3 studies)

The wording matters because the alternative reading is so easy to fall into. "Prey evolved thicker shells to defeat crabs" reads naturally and says something false: it puts a purpose in a process that has none, and it implies the shells arrived in response to a need rather than through variation that already existed being filtered. NatureHQ writes these the long way round — shells varied, thicker-shelled individuals survived crab attacks more often, the population’s average shell thickness rose — because the short version smuggles in the thing this page exists to remove.

Why prey are usually ahead, and when they are not

The two sides are not playing for the same stakes.

What a failure costs each side
AspectPreyPredator
Cost of one failureDeath, and no further offspringA missed meal
Strength of selectionStrong, and immediateWeaker — there will be other attempts
ConsequencePrey tend to be ahead where both investPredators tolerate a low success rate
When it evens outWhere the prey is dangerousA failed attack can injure or kill the predator

The bottom row is where the strongest evidence in this whole area sits. A garter snake attacking a rough-skinned newt is not risking a missed dinner; it is risking tetrodotoxin. That evens the stakes, and it is in exactly this kind of system that reciprocal escalation has been documented most convincingly — toxicity and resistance rising together, and matching each other geographically across many populations.

The geographic matching is the part that makes it evidence. If newts were simply toxic and snakes simply variable, you would expect no particular relationship between local toxicity and local resistance. Finding the most resistant snakes where the most toxic newts live is what a reciprocal process predicts and a one-sided one does not.

Diagram

Trait, counter-trait, and no plan anywhere

Reciprocal selection across generations.

Reciprocal selection — with nothing aiming at anythingPrey populationPredator populationmore toxicmore resistantmore toxicmore resistantmore toxicmore resistantEach side changes because the other changed — over generations.Nobody is responding to anything within a lifetime, and nobody has a plan.
The same explanation in words

Two rows of boxes, prey above and predator below, running left to right across three time steps. The prey boxes read more toxic and the predator boxes more resistant, with arrows running diagonally between the rows: a change in one alters the selective environment of the other, which alters it back. Each side changes because the other changed, over generations. Nothing responds within a lifetime and nothing has a goal — the diagram deliberately shows no direction of intent.

Escalation is not endless

Every trait on both sides has a price, and that is what stops it.

  • Toxins have to be made or acquired and stored safely, which costs energy and carries risk to the animal carrying them.
  • Armour costs mobility, and a slower animal meets a different set of problems.
  • Speed and manoeuvre compete: the body plan that accelerates best is not the one that turns best.
  • Predators face the same accounting, which is why so many break off attacks they might have won.
  • And a prey animal has more than one predator, so a defence that is excellent against one may be irrelevant or costly against another.

Related

  • Predation

    The encounter this operates on

  • Ambush and pursuit

    Matched capabilities, uncertain outcomes

  • Camouflage

    A defence measured against a particular predator’s eyes

  • Venom

    And the resistance that answers it

  • Adaptation

    Why "evolved in order to" is the wrong sentence

The research behind this page

5 studies, newest first. Each one has a page explaining what it found and what it could not show.

2018Nature

Biomechanics of predator–prey arms race in lion, zebra, cheetah and impala

Predators had higher maximum speed, acceleration and deceleration capacity than their prey, but hunts were usually won or lost on manoeuvre rather than on raw performance, and prey succeeded often by turning inside a faster pursuer.

2013Nature

Locomotion dynamics of hunting in wild cheetahs

Cheetahs rarely ran near their maximum speed.

1999BioScience

Predator-prey arms races: asymmetrical selection on predators and prey may be reduced when prey are dangerous

Where prey are dangerous, a failed attack can be fatal for the predator too, and the life–dinner asymmetry weakens.

1979Proceedings of the Royal Society of London B

Arms races between and within species

Selection on prey to escape is typically stronger than selection on predators to catch, because the two sides are not playing for the same stakes — the life–dinner principle.

1977Paleobiology

The Mesozoic marine revolution: evidence from snails, predators and grazers

Shell features associated with predator resistance — thickening, narrow apertures, spines, tighter coiling — increased through the Mesozoic alongside the diversification of shell-crushing and shell-drilling predators.

This page is a stop on a longer route

A guided journey reads several subjects in a deliberate order, with an argument for why one follows another. You can join in the middle.

Where to go from here

Each of these follows from something on this page — a relationship in the evidence, a claim people ask about, or the next mechanism along.

How complete this page is, and what it is still missing

NatureHQ publishes its own gaps. This page is at 36% completeness against what we would call a finished subject, and was last reviewed on 2026-09-03. It carries 4 claims and answers 0 mapped search questions.

  • no research from the last few years is attached — check for newer work
  • more experiments could be explained in plain English
  • no popular claim about this subject has been checked yet
  • The garter snake and newt system carries most of the evidential weight here, which makes the page a demonstration rather than a survey.
  • Escalation and de-escalation over long timescales are treated qualitatively.
  • Plant–herbivore arms races follow the same logic and are covered separately under plant defence.