CAPTCHAs are widely used to protect online services, but they have increasingly become vulnerable to machine learning attacks. In this study, we propose a novel approach to generating adversarial Arabic handwritten CAPTCHAs using five adversarial perturbation techniques: Expectation Over Transformation (EOT) Scaled Gaussian Translation with Channel Shifts Attack (SGTCS) Jacobian-based Saliency Map Attack (JSMA) Immutable Adversarial Noise (IAN) Connectionist Temporal Classification (CTC) We also conduct comprehensive usability and security evaluations of the generated CAPTCHAs to assess their effectiveness. By leveraging recent advancements in machine learning, our approach offers a robust strategy for creating Arabic handwritten CAPTCHAs that enhance security against potential attacks.