The Automated Ultra Vires Vulnerability Detection Framework Integrating DistilBERT and TransLVD

With the improvement of network security level, the traditional black-box scanning techniques show limitations in detecting ultra vires vulnerabilities, especially in complex ultra vires scenarios. This paper proposes a lift vulnerability detection scheme based on DistilBERT, called DistilBERT-TransLVD. This approach uses a pre-trained BERT model to construct a Uniform Resource Locator (URL) classifier based on contextual information. Based on the classification results, a customized lift detection policy extracts unique strings from JavaScript Object Representation (JSON) objects in the request-response content before and after the lift scenario. With advanced string similarity computation and keyword matching techniques, automated lift vulnerability detection is achieved. Experimental results show that the proposed method outperforms the current mainstream techniques for detecting ultra vires vulnerabilities in terms of performance.