Most island-arc magmatism appears to result from the lowering of the melting point of peridotite within the wedge of mantle above subducting slabs owing to the introduction of fluids from the dehydration of subducting oceanic crust(1). Volcanic rocks interpreted to contain a component of melt (not just a fluid) from the subducting slab itself are uncommon, but possible examples have been recognized in the Aleutian islands, Baja California, Patagonia and elsewhere(2-4). The geochemically distinctive rocks from these areas, termed 'adakites', are often associated with subducting plates that are young and warm, and therefore thought to be more prone to melting(5). But the subducting lithosphere in some adakite locations (such as the Aleutian islands) appears to be too old and hence too cold to melt(6,7). This implies either that our interpretation of adakite geochemistry is incorrect, or that our understanding of the tectonic context of adakites is incomplete. Here we present geochemical data from the Kamchatka peninsula and the Aleutian islands that reaffirms the slab-melt interpretation of adakites(2), but in the tectonic context of the exposure to mantle flow around the edge of a torn subducting plate. We conclude that adakites are likely to form whenever the edge of a subducting plate is warmed or ablated by mantle flow. The use of adakites as tracers for such plate geometry may improve our understanding of magma genesis and thermal structure in a variety of subduction-zone environments.