Abstract:

Static defense mechanisms face difficulties addressing dynamic hidden API transgression threats due to limited feature and semantic understanding. Active defense has emerged as an effective approach to enhance network security. This paper proposed an active defense method integrating dynamic semantic sensing and adversarial verification to block API overstepping vulnerability attacks. A dynamic web crawling strategy efficiently obtained page data. This data was combined with a MiniLM model to analyze correlations between response payloads and URLs, enabling payload construction. BERT models were fine-tuned to classify URLs into custom categories. Based on these classifications, a Trans-LVD model performed page similarity analysis to quantify URL similarity levels, identify potential overstepping vulnerabilities, and automate security patching and configuration adjustments. This approach enhanced system adaptability and protection against unknown threats. Experiments were conducted using industry-standard benchmarks to demonstrate the method’s effectiveness in detection accuracy, adaptability, and active defense capabilities.

Key words: deep semantic parsing, active defense, overstepping vulnerabilities, adversarial verification