Abstract. The study proposes a hybrid AI-driven automated testing method for enhancing software quality. It employs defect prediction based on BiLSTM neural networks, test priority calculation via Q-learning, and metamorphic testing for verification in oracle-free environments. The method achieves more effective testing with greater accuracy and cost savings. System components were implemented in Python using PyTorch and OpenAI Gym, with deployment via Docker Swarm in a CI/CD environment. Experiments demonstrated a 50% reduction in testing time, 35-40% improvement in defect detection accuracy, and 60-70% decrease in human involvement compared to manual testing. The proposed approach shows practical applicability in agile environments and CI/CD pipelines.