Recently, machine learning models have demonstrated excellent performance in many tasks. The development of easy-to-use training tools and widely available datasets has further accelerated their adoption. However, widespread adoption of such technologies raises concerns about potential misuse by malicious entities. Furthermore, given the complexities of current ML tools, the lack of documentation can obscure malicious intent within ML models. This project seeks to acquire elements of model transparency and interpretability through forensics analysis of ML models. The project intends to leverage on large pre-trained models to develop automated reverse engineering techniques such as model inversion, so as to gain insight of the model and uncover malicious intents.