CAPTCHA (Completely Automated Public Turing test to tell Computers and Humans Apart) is a widely used security mechanism designed to distinguish human users from automated bots by presenting challenges that are simple for humans but difficult for machines. This study explores the application of convolutional neural networks (CNNs) for the automated recognition and transcription of text-based CAPTCHA images, with the goal of assessing model performance and exposing potential vulnerabilities in traditional CAPTCHA systems. Using a dataset of 1,070 CAPTCHA images, the approach incorporates image preprocessing, segmentation, Synthetic Minority Over-sampling Technique (SMOTE), and data augmentation to enhance model accuracy. The proposed CNN model achieved a classification accuracy of 88%, indicating that contemporary deep learning techniques can effectively compromise conventional CAPTCHA mechanisms. These findings underscore the need for more robust and adaptive security measures in the face of advancing AI capabilities.