Four active threat models
Our work spans configuration security, adversarial machine learning, and the security implications of large language models as they move into finance and healthcare. Each project pairs a real deployment context with a concrete attack surface.
Configuration Hardening for Software Systems
Configuration hardening in software systems is a complex challenge in IT due to the vast number of configuration parameters. Our goal is to develop an AI-powered automated configuration management system that combines efficient configuration hardening with proactive defense strategies against attacks on misconfigured applications.
config: 02Exploring LLM Security in the Finance Sector
LLMs have transformed how we interact with the internet, simplifying customer experiences and automating tasks such as content generation, summarization, and Q&A. But their adoption brings significant security and privacy challenges. Our work examines these implications within the financial sector — exploring how LLMs can be used securely and ethically in banking to enhance trust and resilience.
config: 03Exploring LLM Security in Medical Applications
AI has revolutionized the use of Electronic Health Records (EHR) by enhancing efficiency, accuracy, and patient care, and LLMs are now transforming EHR interaction itself — enabling more intuitive workflows and reduced documentation burden. We study the privacy and security attacks that become possible once sensitive EHR data flows through an LLM.
config: 04Biometric User Authentication Security in Online Banking
Traditional password-based authentication is inefficient and prone to attack. Real-time facial recognition offers a secure alternative by ensuring only authorized users perform sensitive transactions — but adversarial machine learning (AML) attacks can reverse-engineer raw biometric data to mimic recognition models and enable unauthorized access. We study face recognition and verification methods in online banking alongside the attack and defense techniques that protect them.