Adaptive Threat Mitigation in 6G Network Slicing: A Machine Learning Approach to Dynamic Security Policy Orchestration

Background: With the listing of 6G network slice architecture, you need to have a very advanced security mechanism that can cope with the very fast-evolving threat landscape as well as ensure that low reaction time takes place. The old static security frameworks prove insufficient for 6G's rapidly evolving, heterogeneous environment. Objective: This research develops and evaluates an adaptive threat mitigation system that uses machine learning for dynamic security policy orchestration across network slices. Methods: We propose AMSTM (Adaptive ML-based Security Threat Mitigation), which combines Deep Deterministic Policy Gradient reinforcement learning and Graph Attention Networks. The model was evaluated in two functional scenarios(OMNeT++ with 6G, the specific environment) for 20 slices of different matrices and 15 distinct attack settings. Results: As against other systems, AMSTM got 94.7% threat detection accuracy, cut the number of warning information by 67% and reduced the average response time to critical warning events to 12.3ms. It also achieved a 97.3% containment efficiency across slices under attack condition 1 or 2, while there is still less than 0.25 ms ULRLC-latency growth with the slice under attack on other characteristics. Conclusions: The united model delivers adaptive security orchestration that can serve the onset of 6G, offering logarithmic scaling and up to 1,000 slices at one time.