Honeypots deceptive systems designed to attract and monitor malicious activity have long served as valuable tools in cybersecurity. Traditional honeypots demonstrate their usefulness yet their fixed configurations and inactive responses restrict their ability to combat complex evolving cyber threats. The research presents an ML-powered honeypot system which integrates with AWS to enhance detection capabilities and automated response functions. The system employs both anomaly detection algorithms together with reinforcement learning models to modify its operations in real time through attacker interaction. Performance metrics demonstrate substantial advancements. The system demonstrates an anomaly detection accuracy rate exceeding 95% while responding within 500 ms and producing less than 2% false positives. Reinforcement learning techniques enhanced detection precision by 30% compared to conventional honeypots specifically for adaptive threats. The research presents both architectural framework and ML model development and AWS service integration for the system. The presented method provides an actionable framework for intelligent proactive defense against cyber threats. The research demonstrates how adaptive honeypots decrease modern cybersecurity threats while generating valuable knowledge for developing autonomous threat management systems.