Research

Research Overview

CITCS at Texas A&M University-San Antonio is committed to advancing innovative research that addresses the evolving challenges of the digital world. Our faculty lead impactful basic, applied, and interdisciplinary research initiatives that strengthen the security, resilience, and trustworthiness of modern computing systems and critical infrastructure. Through high-quality scholarly publications, externally funded projects, and collaborations with industry and government partners, our researchers continue to contribute meaningful advancements to the cybersecurity and AI fields.

Our research spans a broad range of areas, including secure software engineering, artificial intelligence security, cyber defense, digital forensics, network and cloud security, privacy preservation, critical infrastructure protection, post-quantum cryptography, and cyber workforce development. By integrating research with education and community engagement, the center prepares the next generation of cybersecurity and AI professionals while contributing to the protection of local, national, and global cyber and AI ecosystems.

Defending AI Coding Models against Data Poisoning Attacks

Code-Centric Models (CCM) and Large Language Models (LLM)-based vulnerability detection systems are vulnerable to backdoor data poisoning attacks, leading to hidden and targeted misclassification. To address the gap, this research aims to assess the security of Rust code generated by CCMs and LLMs under targeted data poisoning attacks and to design a defense framework that can detect poisoned code samples and remove them from datasets used in model training and deployment. This work involves three tasks: Task 1 benchmarks comprehensive Rust Datasets suitable for training and fine-tuning models; Task 2 simulates backdoor data poisoning attacks with code triggering strategies under threat models deployed; Task 3 designs a defense framework to detect triggered code samples and remove them from datasets used during model training, fine-tuning, and deployment.

attack-pipeline

CCPG-Rust: A Multi-Tool Exchange Unified Framework for Comprehensive Rust Vulnerability Detection using Concurrent Code Property Graph

This research aims to develop a first noble exchange framework, CCPG-Rust, which integrates symbolic execution traces, model checker counter examples, and dynamic debugger observations into a single unified Concurrent Code Property Graph (CCPG), enabling large-scale and comprehensive cross-tool vulnerability detection in Rust. This framework represents a transformative approach in program analysis, moving from isolated outputs produced by the above-mentioned different tools and approaches to a unified synthesis of multi-source evidence. The architecture pipeline of the framework involves five high-level steps (Fig. 1): (1) Rust Code Input, (2) Multi-Tool Analysis, (3) Integration to Graph Database, (4) Synthesis and Detection, and (5) Interactive Output Interface.

CCPG