Predictive Threat Modeling Using Reinforcement Learning Agents for API Gateway Exploit Detection
Main Article Content
Abstract
Using APIs so widely in current digital systems has made them more vulnerable to a wide range of modern cyber attacks. Standard security systems such as signaturebased detection and firewalls, cannot protect against adaptable, hidden and new attacks on API vulnerabilities. The research team created a new way to model threats, relying on RL agents and Deep Q-Learning, to learn how to respond to API gateway attacks. We come up with a distributed framework that involves multi-agent collaboration, so that it can detect injection attacks, navigate authentication methods and block the patterns used in distributed denial-of-service (DDoS). We discovered that by applying our model, detection accuracy raised, Defense against cyber-threats was improved and responses were provided more quickly. AI does its part by also making use of explainable AI to give more insight and meet the requirements for compliance. From the data results, our model is seen as more capable than traditional ML and deep learning in achieving a better F1-score, FPR and reducing the time required for inference. We close by indicating several future goals, especially using federated learning and energy-saving training methods for environments with limited resources. What we have done underlines the value of reinforcement learning agents in threat modeling and marks a practical way to advance security in API-driven systems.