Stress intensity factor is a crucial parameter for modeling and predicting structural fracture failure. This study evaluates the dynamic stress intensity factor for solving three-dimensional dynamic fracture problems using the adaptive phantom node method. This technique combines the phantom node method with adaptive mesh refinement, automating the generation of a dense mesh around the crack. In this approach, strong discontinuities at cracks are modeled using phantom nodes without crack tip enrichment functions or extra degrees of freedom. The theoretical framework of this technique is straightforward and easy to implement based on the finite element method, but it requires a relatively dense mesh to ensure computational accuracy. Adaptive mesh refinement technology and criteria suitable for crack problems are introduced into the phantom node method, thus obviating the need for a globally dense mesh with high computational consumption while improving computational accuracy and efficiency. A concise approach, known as constrained approximation, is adopted to deal with hanging nodes presented in the locally refined mesh. It is convenient to implement numerically, does not involve special elements or complex shape functions, and retains the interpolation and numerical integration of the standard finite element method. The stress intensity factors for several three-dimensional crack problems are evaluated using the adaptive phantom node method and compared with the theoretical solutions and numerical results obtained by the standard phantom node method. It is found that the numerical results of this method are in good agreement with the theoretical solutions, and the computational accuracy is effectively improved compared to the standard phantom node method. Additionally, compared to the locally pre-refined mesh with equivalent accuracy, the adaptive refined mesh exhibits higher computational efficiency and reduced computational consumption. This holds considerable potential value for the efficient simulation and prediction of dynamic fracture failure in large-scale complex engineering structures.