Research Projects

Large Language Models (LLMs) such as GPT, BERT, and LLaMA are revolutionizing cybersecurity with their ability to process, understand, and generate human-like text. These models offer significant potential in threat detection, phishing prevention, automated incident response, and cyber threat intelligence. However, their capabilities can also be exploited for malicious purposes, including phishing attacks, misinformation, data leakage, and adversarial manipulation. This project, titled LLMShield, seeks to explore the dual nature of LLMs in cybersecurity: as a powerful tool for defense and a potential vector for sophisticated attacks. We aim to develop an autonomous, secure, and explainable cyber threat intelligence framework that leverages LLMs while implementing novel mechanisms to prevent misuse, safeguard privacy, and resist adversarial exploitation. Scientific Rationale: After a detailed study, the following scientific rationales come out as the basis of this proposal. (i) LLMs as a Double-Edged Sword - Opportunities: Effective in phishing detection, malware analysis, and cyber threat intelligence (Ferrag et al., 2024; Microsoft Security Copilot, 2023). - Threats: Misused for phishing, malicious code generation, data leakage, and adversarial prompt injection (Perez et al., 2022; Carlini et al., 2021). (ii) Research Gaps - Current studies address either applications or attack surfaces, rarely an integrated defense-attack framework. - Most defenses (Chan, 2025; Wang, 2025) are lab-scale with limited real-world deployment. - India lacks indigenous research on LLM-powered autonomous threat intelligence with built-in security mechanisms. (iii) Scientific Motivation - Growing AI-driven cyberattacks and misinformation campaigns necessitate robust LLM security. - Combining attack surface analysis, defense engineering, and autonomous cyber threat intelligence is a unique and timely research need. (iv) Hypothesis “Properly secured LLMs can serve as autonomous, explainable, and scalable tools for reliable cyber threat intelligence.” (v) Research Contribution - Investigating LLM-specific attack vectors and proposing defenses. - Building a deployable LLM-powered Threat Intelligence System (LLM-TIS). - Proposing policy recommendations for safe LLM Novelty of the Proposal: (i) First Indian initiative to combine attack analysis, defense engineering, and autonomous cyber threat intelligence in a single framework. (ii) Introduction of LLM-powered Threat Intelligence System (LLM-TIS) capable of real-time, explainable threat correlation. (iii) Generation of policy recommendations for safe LLM adoption in critical infrastructure. Importance of the Proposal: (i) Addresses an urgent national security need by mitigating risks of AI-driven cyberattacks. (ii) Builds indigenous capacity in securing and deploying LLMs for cyber defense. (iii) Contributes new knowledge, open-source tools, and patents, strengthening India’s leadership in AI-driven cyber resilience.