Research indexRCT
Breaking Secure CAPTCHA using Deep Learning
Khan, Ahmed, Noreen and 2 others
2 Jun 2025Open access
NobleIDark:/48914/w1/20261105/ED789FB7DOI10.67535/tsp.000001.013
Breaking Secure CAPTCHA using Deep Learning
Khan, Ahmed, Noreen and 2 others · 2025
NobleIDark:/48914/w1/20261105/ED789FB7DOI10.67535/tsp.000001.013
Abstract
The cornerstone of web safety, CAPTCHA (Completely Automatic Public Turing test to tell the Computer humans apart), CAPTCHA systems are increasingly vulnerable to deep learning and computer vision-based attacks. This study illustrates a vulnerability in text-based CAPTCHA using a Convolutional Recurrent Neural Network (CRNN) framework that automates solutions with 99.52 percent accuracy. The traditional immune system, overlapping signs, and background noise are systematically overcome by combining CNNs (for spatial feature extraction) with RNNs (for sequential decoding). To address the lack of data, we fine-tune the model using a publicly available dataset from Kaggle of synthetic CAPTCHA images with sinus-type distortion, Gaussian noise, and randomized fonts. The model achieves 99.52 Percent accuracy for various variations (strike, blockage, multicolored backgrounds) with a 1.5-second inference time for consumer GPUs, highlighting the practical feasibility of adversarial attacks. It exceeds 25 percent of independent CNN/RNN, indicating insufficient static, distortion-dependent design. Generalizing invisible fonts and distortion levels has confirmed that visual veils alone cannot prevent AI. In addition to technical contributions, a framework is available to catalyze AI resistance certification research. Alternative dynamic adaptive visual fantasies, behavioral biometry, or logic puzzles are used to explore human-machine cognitive gaps. CAPTCHA reproducibility tools for stress testing have been proposed, addressing human-centered learning tolerance security in an age controlled by AI.