Abstract: Agricultural irrigation in loess tablelands is shifting from traditional flood irrigation to water-saving irrigation. However, under the high groundwater levels formed by long-term flood irrigation, the mechanisms by which this transition affects internal moisture redistribution and water-induced deterioration of deep soil remain unclear. Focusing on a typical fractured loess slope in Heifangtai, Gansu Province, this study establishes a coupled seepage-deformation numerical model with intermittent infiltration boundary conditions. The spatiotemporal evolution of the seepage field and slope stability under flood and drip irrigation modes is systematically investigated. The results show that: (1) internal moisture seepage exhibits significant spatiotemporal variability. Flood irrigation readily induces preferential flow along fractures, rapidly reducing effective soil suction. By contrast, drip irrigation is dominated by slow matric flow, and the effective suction of deep soil dissipates gradually under the buffering effect of the loess vadose zone. (2) Water-induced slope deterioration and instability exhibit a pronounced time-lag effect. Even during drip-irrigation intervals, continuous downward redistribution of retained moisture causes a delayed but abrupt drop in effective suction at the slope toe, degrading soil shear strength and ultimately triggering progressive tension-shear composite failure. (3) Deep groundwater level is the critical factor limiting the hazard-mitigation benefit of drip irrigation. Over a complete hydrological year (365 days), a groundwater level of ≤ 12.5 m allows the deep unsaturated zone to retard downward wetting-front migration and maintain slope stability. When the groundwater level is >= 15.0 m, however, the descending wetting front rapidly couples with the capillary fringe, sharply increasing slope-instability risk; a further rise in groundwater level may directly trigger slope failure. For hazardous slopes controlled by high groundwater levels and a pre-existing weak base, disaster-specific and classified prevention measures are recommended to minimize risk.