Breast cancer is the most common cancer diagnosed worldwide, and mammography remains the gold-standard modality for screening. Nowadays, Artificial Intelligence (AI) is a valuable tool, helping clinicians reach their diagnosis with more confidence and in less time. However, a significant issue arises with the increasing use of AI: validation. Since AI outputs can be sensitive to acquisition settings, controlled and reproducible test objects are needed to assess their robustness. This thesis presents the development of a 3D-printed breast phantom for AI validation. A fully automated Python pipeline was designed to convert raw DICOM mammograms into STL files ready to be printed, using an inverse attenuation model to translate the raw intensity data into height maps.