The Corporate Social Responsibility (CSR) reports are crucial for communicating a company's sustainability and ethical practices, but manual analysis of these reports is often inefficient and prone to bias. This study aims to develop an automated framework for the thematic analysis of CSR reports using advanced Natural Language Processing (NLP) and machine learning techniques. The methodology involves generating semantically rich text embeddings using BERT, clustering these embeddings into thematic groups with KMeans, and employing Principal Component Analysis (PCA) and word clouds for dimensionality reduction and visualization. The results demonstrate high-quality clustering, with a Silhouette Score of 0.65, and identify key themes such as Sustainability and Governance. Insights reveal patterns in CSR practices across various industries and regions, highlighting the multifaceted nature of modern corporate responsibility. This study contributes to the field by enhancing the scalability and efficiency of CSR analysis, providing actionable insights for stakeholders, and advancing the automation of thematic analysis in corporate reporting.