<p>Deep neural networks (DNNs) have demonstrated remarkable performance across various domains but remain susceptible to adversarial examples, which exploit model vulnerabilities to induce misclassification. While adversarial attacks on static images have been extensively studied, their application to video data poses unique challenges due to the sequential and high-dimensional nature of videos, as well as the need for real-time perturbation generation. To address this, we propose dual-targeted adversarial examples (DTAEs), a novel framework designed to manipulate outcomes by ensuring that two different models misclassify the same input into distinct incorrect classes. Utilizing the I3D model and the CNN+LSTM model as target models, we generate adversarial perturbations that lead each model to predict a different erroneous class using the Kinetics-400 dataset. Experimental results show that when the epsilon is set to 0.07, the proposed adversarial examples achieve a 98.0% attack success rate for the I3D model and a 97.5% attack success rate for the CNN+LSTM model, with each model misclassifying the inputs into different incorrect classes. This work highlights the feasibility of dual-targeted adversarial attacks on video recognition systems and underscores the need for robust defense mechanisms.</p>