Subduction zones are essential drivers of plate tectonics. However, the processes causing subduction zone initiation (SZI), involving the formation of a new plate boundary, and the forces required remain disputed. Here, we focus on horizontally forced SZI at passive margins and quantify the horizontal force required for SZI with two-dimensional petrological-thermomechanical numerical models. The initial configuration, involving two passive margins bounding a 400 km wide basin with exhumed mantle, is calculated by a numerical rifting simulation to guarantee thermomechanical feasibility and isostatic equilibrium. SZI occurs at one passive margin during convergence of the basin-margin system. SZI is caused by thermal softening due to local shear heating. A local temperature increase of only ca. 50°C is sufficient to cause SZI. Corresponding simulations without shear heating do not show SZI, showing unequivocally that thermal softening controls SZI. We systematically limit the lithospheric deviatoric stress to determine the minimum force required for SZI. A minimum force (per unit length) of ca. 14 TN m−1 is required for SZI; a force magnitude that agrees with independent estimates from mantle convection models. However, for the associated stress limit of 200 MPa the lithosphere is so weak that slab detachment (SD) already occurs at the onset of basin closure in the horizontal, non-subducted region of the subducting plate. Only for stress limits >=300 MPa (yielding a force of ca. 15 TN m−1) subduction of continental crust occurs and SD occurs in the subducted slab. The force required for SZI is a proxy for the effective compressive lithospheric strength. This strength also controls subsequent SD due to tensile slab pull, which increases during subduction, also since our models include the olivine–wadsleyite phase change. Our simulations show a causal link betweeen SZI and subsequent SD: If forces required for SZI are smaller, then the lithosphere is weaker and then SD occurs at shallower levels and corresponding slabs are shorter. Concerning the European Alps, our results imply that if there was no SD since the Europe–Adria collision some 30 Myr ago (as proposed by some studies), then the subducted lithosphere must be considerably strong.