Various platforms and web apps to deliver material via the Internet are becoming more widespread. Web-based technologies accept and store sensitive information from users. Because of their Internet connectivity, these systems and the databases they link to are vulnerable to various information security vulnerabilities. The most dangerous threats are denial of service (DoS) and SQL injection assaults. SQL Injection attacks are at the top of the list for web-based systems. In this type of attack, the perpetrator will take sensitive and classified information that might hurt a firm or enterprise. Depending on the conditions, the corporation may incur financial losses, have private information disclosed, and have its stock market value drop. This work uses machine learning-based classifiers such as MLP, Support Vector Machine, Logistic Regression, Naive Bayes, and Decision Tree to identify and detect SQL Injection attacks. For the SQLI dataset learning strategies, we examined all five algorithms using the Confusion Matrix, F1 Score, and Log Loss. We discuss the benefits and drawbacks of the proposed AI-based SQLI techniques. Finally, we talk about making SQLI reach its full potential through more research in the coming years.

Machine learning and deep learning are widely utilized and highly effective in attack classifiers. Little research has been undertaken on detecting and protecting cross-site scripting, leaving artificial intelligence systems susceptible to adversarial assaults (XSS). It is crucial to develop a mechanism for increasing the algorithm's resilience to assault. This study intends to utilize reinforcement learning to enhance XSS detection and adversarial combat attacks. Before mining the detection model's hostile inputs, the model's information is extracted using a reinforcement learning framework. Second, the detection technique is simultaneously trained using an adversarial strategy. Every cycle, the classification method is educated with freshly discovered harmful data. The proposed XSS model effectively mines destructive inputs missed by either black-box or white-box detection systems during the experimental phase. It is possible to train assault and detection models to enhance their capacity to protect themselves, leading to a lower rate of escape due to this training.