Abstract: This paper presents a Bayesian physics-informed neural network (B–PINN) prediction framework, developed based on the phase field fracture model, to quantify the impact of reservoir uncertainty on hydraulic fracture propagation. The proposed method realizes the coupled modeling of the mechanical, seepage, and phase fields by embedding governing physical constraints into the loss function. Simultaneously, Bayesian inference is employed to effectively capture the output uncertainty distribution, thereby offering quantitative risk assessment metrics for fracture design. Numerical results demonstrate that, compared with the finite element solution of the phase field hydraulic fracturing model implemented in COMSOL, the B–PINN achieves phase field prediction errors within 8.7% and 12.3% in single- and double-fracture propagation scenarios, respectively. Additionally, the computational efficiency is enhanced by approximately 50% under GPU acceleration. This study provides a solution that is both theoretically rigorous and practically efficient, offering strong support for the robust optimization of hydraulic fracturing designs in the presence of complex geological uncertainties.