AbstractCratons are generally observed to retain thick (>180 km) conductive keels for billions of years. However, some cratons have undergone keel removal, with well‐documented examples being the eastern North China Craton (NCC) and the Wyoming Craton (WC). These keelless subregions appear to have kept a lithospheric bottom at ~80–100‐km depths. This is also the depth range where modern cratons, including the remaining portions of the NCC and the WC, have seismically visible midlithospheric discontinuity layers (MLDLs). MLDLs are proposed to be regions of preferential accumulation of metasomatic minerals and/or anomalously wet (>1,000 ppm) peridotites, both of which would lead to a relatively weak rheology. We propose that the cratonic keels of the eastern NCC (ENCC) and the western WC (WWC) utilized this weak MLDL layer to delaminate from overlying lithosphere. We first explore this hypothesis with a lubrication‐theory‐based analytical model. This model suggests a close relationship between a cratonic keel's long‐term stability and the strength of the MLDL's edge. We further test this prediction with less idealized 2‐D numerical experiments which reveal that (a) dense lower keels beneath MLDL‐bearing cratons can persist for billions of years as long as the MLDL's edges abut relatively cold and strong lithosphere; (b) MLDL edge failure can induce rapid intramantle lower keel delamination; and (c) the predicted rates of keel delamination along a ~10‐km‐thick MLDL with a hydrous olivine or metasomatic mineral‐dominated rheology are consistent with observations for the removal speeds of the WWC and the ENCC.