The updip limit of the seismogenic zone of megathrusts is poorly understood. The relative absence of observed microseismicity in such regions, together with laboratory studies of friction, suggests that the shallow fault is mostly velocity strengthening, and likely to creep. Inversions of geodetic data commonly show low to zero coupling at the trench, reinforcing this view. We show that the locked, downdip portion of the megathrust creates an updip stress shadow that prevents the shallow portion of the fault from creeping at a significant rate, regardless of its frictional behavior. Our models demonstrate that even if the shallowest 40% of the fault is frictionally unlocked, the expected creep at the fault tip is at most 30% of the plate rate, often within the uncertainties of surface geodetic measurements, and below current resolution of seafloor measurements. We conclude that many geodetic models significantly underestimate the degree of shallow coupling on megathrusts, and thus seismic and tsunami hazard.