"Citadel Researchers Propose A Deep Learning Technique To Generate DNS Amplification Attacks"
Deep learning algorithms have shown promise in detecting and characterizing cybersecurity breaches. However, fraudsters have been working on new attacks to disrupt the operation of various deep learning systems, such as those used for image analysis and natural language processing. These attacks may cause the failure of many applications, biometric systems, and other systems involving deep learning algorithms. Previous studies have demonstrated the effectiveness of various adversarial approaches in causing Deep Neural Networks (DNNs) to make untrustworthy and inaccurate predictions. Citadel researchers recently built a DNN capable of detecting DNS amplification, a type of Denial-of-Service (DoS) attack. Then they used two different techniques to create adversarial samples capable of fooling their DNN. Their findings show that deep learning approaches for detecting DNS intrusions are inaccurate and vulnerable to adversarial attacks. Distributed DoS (DDoS) DNS amplification attacks exploit vulnerabilities in DNS servers to amplify requests sent to them, eventually flooding them with data and bringing them down. These attacks have the potential to severely disrupt Internet services provided by both large and small multinational corporations. Deep learning algorithms for detecting DDoS DNS amplification attacks have been developed in recent years by computer scientists, but the Citadel team demonstrated that those algorithms could be avoided through the use of adversarial networks. Elastic-Net Attack (EAD) and TextAttack are techniques that have been demonstrated to be effective in producing corrupted data that DNNs would misclassify. In their experiments, the team discovered that the TextAttack algorithm could generate adversarial examples with a 100 percent chance of deception against the model, and the adversarial examples from the EAD algorithm had a 67.63 percent chance of deceiving the model. This article continues to discuss the new study showing how deep learning models trained for network intrusion detection can be circumvented.