With the rapid development of network technology, cyber threats have become increasingly sophisticated, posing significant risks to secure communications and privacy preservation. Under advanced persistent threats (APTs) and emerging quantum computing-based attacks, traditional passive defense mechanisms are often powerless in detecting and resisting attacks in real-time. To address these challenges, we propose an active defense scheme. It integrates traffic anomaly detection with dynamic key updates, enhancing network security while resisting quantum threats. Specifically, we use Convolutional Neural Network (CNN) to analyze real-time network traffic, identifying potential anomalies. If abnormal traffic is detected, the scheme will trigger the key update by using Kyber. The updated key is then utilized for subsequent network communication. We evaluate the proposed scheme on the UNSW_NB15 and NSL-KDD datasets. Experimental results demonstrate that the proposed scheme achieves a low detection latency and fast key update latency across different network conditions. Also, the system maintains a stable throughput and exhibits minimal CPU usage, ensuring robust network security without significantly compromising performance.