Peer-Reviewed Journal Articles: 3 (including 1 accepted for publication)
Accepted Conference Papers: 1
Standalone Preprints: 4
Research Areas: Applied Cybersecurity, AI Security, Secure Software Engineering, Deepfake Detection, Trustworthy AI, Secure Systems
My research focuses on applied cybersecurity, AI-assisted software development security, software vulnerability analysis, deepfake detection, and trustworthy AI systems. I investigate how vulnerabilities emerge and evolve in AI-generated software, how automated systems can introduce cumulative security risks, and how evidence validation and multimedia analysis can strengthen trust in AI-enabled technologies.
My work combines empirical cybersecurity research, machine learning, program analysis, and experimental evaluation to develop practical approaches for improving software reliability, digital trust, and secure systems design.
This page provides access to my peer-reviewed journal articles, accepted conference research, preprints, and publicly available research artifacts. Preprints are listed separately from peer-reviewed publications, and multiple repository versions of the same work are counted only once.
Peer-Reviewed Journal Articles
Hybrid Classification for Complex and Composite Activity Recognition in Multi-User Environments
Halil Ibrahim Dursunoglu (2026)
IEEE Sensors Journal — Accepted for publication
DOI: 10.1109/JSEN.2026.3741056
This research presents a hybrid machine learning framework for recognizing complex and composite human activities in multi-user environments. The study examines activity classification, contextual relationships, and individual attribution using wearable sensor data, contributing to intelligent sensing, machine learning, and human activity recognition.
Audio-Visual Synchronization Analysis for Deepfake Detection: A Comprehensive Review
Halil Ibrahim Dursunoglu (2026)
International Journal of Computer Techniques (IJCT)
Repository DOI: 10.5281/zenodo.19557051
This review examines audio-visual synchronization approaches for detecting manipulated and AI-generated media. It discusses temporal inconsistencies between speech and facial movements, synchronization-based detection methods, evaluation challenges, and emerging threats involving AI-generated impersonation, multimedia manipulation, and digital media authenticity.
Human Activity Monitoring with Wearable Sensors and Hybrid Classifiers
G. Uslu, H. I. Dursunoglu, O. Altun, S. Baydere (2013)
International Journal of Computer Information Systems and Industrial Management
This study investigates human activity recognition using wearable sensor data and hybrid classification techniques. It explores machine learning approaches for identifying physical activities and contributes to intelligent monitoring and sensor-based activity analysis.
Peer-Reviewed Conference Papers
SecurePromptBench: Benchmarking Security Retention in Iterative AI-Assisted Software Development
Halil Ibrahim Dursunoglu (2026)
Sixth Intelligent Cybersecurity Conference (ICSC 2026)
Status: Accepted and presented; proceedings forthcoming.
This research investigates how software security is maintained or degraded across iterative AI-assisted development workflows. It examines vulnerability propagation, security retention, and the risks associated with repeated LLM-generated code modifications, emphasizing the importance of longitudinal security evaluation rather than relying exclusively on one-time vulnerability assessments.
Preprints & Working Papers
The following manuscripts represent publicly available research contributions that have not been listed here as peer-reviewed publications.
SEMA-GUARD: Semantic and Graph-Based Vulnerability Detection in Assembly Code
Halil Ibrahim Dursunoglu, Kaan Sulkalar (2026)
arXiv Preprint — arXiv:2609.17254
arXiv DOI: 10.48550/arXiv.2609.17254
Full Paper: View on arXiv
Zenodo Version: 10.5281/zenodo.20550170
SEMA-GUARD introduces a semantic and graph-based vulnerability detection framework for assembly language and low-level software systems. The approach combines graph neural networks with execution-level semantic representations, including control flow, stack operations, memory access patterns, and data-flow relationships.
Evaluated using assembly programs derived from the Juliet Test Suite, the framework reports 85.1% classification accuracy and an F1 score of 0.801, demonstrating the potential of semantic graph representations for identifying software vulnerabilities at the assembly level.
Security Drift in AI-Assisted Software Development: A Framework for Measuring the Security Impact of Iterative Prompt Evolution
Halil Ibrahim Dursunoglu, Kaan Sulkalar (2026)
SSRN Preprint — June 25, 2026
DOI: 10.2139/ssrn.7467762
Full Paper: View on SSRN
This research introduces a framework for examining security drift during iterative AI-assisted software development. It investigates how repeated prompt modifications and code refinement can influence software security over time, emphasizing longitudinal vulnerability assessment and the need to measure cumulative security changes across development iterations.
Security Weaknesses in LLM-Generated Source Code: An Empirical Vulnerability Analysis of Iterative AI-Assisted Development
Halil Ibrahim Dursunoglu, Kaan Sulkalar (2026)
Research Preprint
Repository DOI: 10.5281/zenodo.20550245
This empirical study investigates how security vulnerabilities evolve throughout iterative large language model-assisted software development. It examines vulnerability introduction, persistence, and propagation across repeated code modifications and evaluates approaches for identifying and measuring security degradation throughout AI-assisted development workflows.
CiteGuard-RAG: A Validation-Centered AI System for Evidence-Grounded Question Answering
Sumit Barua, Guan Hong, Halil Dursunoglu, Charles Rodgers, Alvis Fong (2026)
arXiv Preprint — arXiv:2609.15830
arXiv DOI: 10.48550/arXiv.2609.15830
Full Paper: View on arXiv
CiteGuard-RAG presents a validation-centered AI framework for improving the reliability of retrieval-augmented question answering. The system combines semantic and lexical retrieval, citation-constrained generation, sentence-level evidence validation, and controlled regeneration to reduce unsupported claims and improve evidence grounding.
Evaluated across 400 questions from multiple datasets, the research explores how evidence validation, citation verification, and calibrated refusal mechanisms can support more trustworthy AI-generated responses.
Research Software & Open Research Artifacts
Reproducibility, transparent experimental evaluation, and practical implementation are central to my research approach. Where available, I share software prototypes, experimental frameworks, benchmark datasets, evaluation scripts, and supporting research materials through public repositories.
These resources support independent examination, replication, and continued research in AI-assisted software security, vulnerability detection, trustworthy AI, and secure systems engineering.
Research Areas
AI-Assisted Software Security
Security of LLM-generated source code, vulnerability propagation, iterative prompt evolution, security drift, secure coding practices, and longitudinal software security evaluation.
Software Vulnerability Detection & Program Analysis
Static and dynamic security analysis, semantic vulnerability detection, graph-based program representations, assembly-level security analysis, and automated vulnerability assessment.
Deepfake Detection & Digital Trust
Audio-visual synchronization, AI-generated impersonation, multimedia authenticity, deepfake detection, and digital media forensics.
Trustworthy AI & Evidence-Grounded Systems
Retrieval-augmented generation, evidence validation, citation verification, AI reliability, and reducing unsupported AI-generated responses.
Secure Systems & Automated Infrastructure
AI-driven network automation, configuration security, security policy validation, security drift in automated infrastructure, and resilient systems architecture.
Applied Machine Learning & Intelligent Sensing
Human activity recognition, hybrid classification, wearable sensor analysis, intelligent monitoring, and machine learning-based systems evaluation.
Editorial & Scholarly Service
I serve as an Editor for Cyber Serum Journal, a peer reviewer for IEEE Access, IEEE Transactions on Multimedia, Journal of Computer Science, Journal of Open Source Software (JOSS), and Journal of Open Research Software (JORS), and a Technical Program Committee member for the International Conference on Cybersecurity Systems, Technologies and Applications (CSTA 2026).
My scholarly service focuses on evaluating research quality, methodological rigor, technical implementation, experimental validation, reproducibility, and the practical implications of emerging technologies in cybersecurity, artificial intelligence, and software engineering.
GitHub
Source code, software prototypes, educational security labs, and research implementations.
Zenodo
Publications, preprints, datasets, and archived research artifacts with DOI assignment.
Google Scholar
Citation records, publication metrics, and indexing information.
ORCID
Persistent researcher identifier and publication record.