MUMBAI, India, Aug. 10 -- Intellectual Property India has published a patent application (202641091567 A) filed by Mrs. Khasimbee Shaik; Dr. K. V. Satyanarayana; and Dr. Tirimula Rao Benala on July 28, 2026, for System And Method For Detecting Malicious Code Injection Using Semantic Embeddings And Gradient Q-Learning Networks.

Inventors include Mrs. Khasimbee Shaik; Dr. K. V. Satyanarayana; and Dr. Tirimula Rao Benala.

The application for the patent was published on July 31, 2026, under issue no. 31/2026.

Abstract: ABSTRACT SYSTEM AND METHOD FOR DETECTING MALICIOUS CODE INJECTION USING SEMANTIC EMBEDDINGS AND GRADIENT Q- LEARNING NETWORKS The invention discloses a system and method for detecting malicious code injection by integrating semantic embeddings derived from large language models with gradient-based reinforcement learning. The framework employs LLMVec2 Sentence-BERT embeddings to transform source code into contextual vector representations that capture syntactic and semantic relationships between program tokens. Convolutional layers extract local structural features, while a Gradient Regression Vector module models behavioral deviations in embedded code representations. A Q-learning classifier dynamically optimizes decision boundaries, treating each embedded representation as a state, classification outcomes as actions, and detection accuracy as reward signals. The end-to-end pipeline begins with dataset ingestion, normalization, and tokenization of benign and malicious code samples, followed by embedding generation, convolutional feature extraction, gradient regression analysis, and reinforcement-based classification. Experimental evaluation on benchmark datasets demonstrates superior performance compared to conventional machine learning and deep learning baselines, achieving 95% accuracy, 94% precision, 95% recall, and 95% F1-score, with RMSE reduced to 0.085. The invention provides a scalable, language-agnostic, and semantically aware detection system resilient against obfuscated and multi-line injections, thereby enhancing the security and reliability of modern software development environments.

Disclaimer: Curated by HT Syndication.