My research focuses on applied cybersecurity, artificial intelligence security, secure software engineering, and trustworthy AI-enabled systems. I investigate how emerging technologies introduce new security risks, particularly in AI-assisted software development, automated infrastructure, and AI-generated digital media.
A central theme of my work is understanding how security weaknesses emerge, evolve, and propagate over time. Rather than evaluating software or AI systems only at a single point, I am interested in developing methods for continuous security assessment, vulnerability detection, longitudinal security evaluation, and evidence-based validation.
Drawing on experience in academic research, software engineering, and cybersecurity education, I aim to bridge theoretical advances with practical approaches that improve software reliability, strengthen digital trust, and enhance the resilience of modern computing systems.
1. AI-Assisted Software Security & Security Drift
Large language models are increasingly used to generate, debug, refactor, and maintain software. Although these capabilities can accelerate development, they also introduce security challenges that may not be apparent during initial code generation.
My research examines how vulnerabilities emerge, persist, and accumulate throughout iterative AI-assisted software development. A particular focus is security drift, where repeated prompts, code modifications, and automated development decisions gradually change the security posture of an application.
This research investigates longitudinal vulnerability measurement, security retention, vulnerability propagation, and approaches to continuous security validation.
Selected Research
Security Drift in AI-Assisted Software Development: A Framework for Measuring the Security Impact of Iterative Prompt Evolution
Dursunoglu & Sulkalar (2026) — SSRN Preprint
View Paper
Security Weaknesses in LLM-Generated Source Code: An Empirical Vulnerability Analysis of Iterative AI-Assisted Development
Dursunoglu & Sulkalar (2026) — Research Preprint
View Paper
SecurePromptBench: Benchmarking Security Retention in Iterative AI-Assisted Software Development
Sixth Intelligent Cybersecurity Conference (ICSC 2026) — Accepted and Presented
Research Objectives
- Understand how vulnerabilities evolve across iterative AI-assisted development.
- Develop methods for measuring cumulative security degradation.
- Investigate vulnerability persistence and propagation across code revisions.
- Improve security validation throughout AI-assisted software development workflows.
2. Software Vulnerability Detection & Secure Systems Engineering
Software security depends on identifying weaknesses not only in high-level source code but also in lower-level program representations and system behavior.
My research explores semantic vulnerability detection, graph-based program analysis, secure coding practices, and automated security assessment. I am particularly interested in combining program semantics with machine learning to identify vulnerabilities that may be difficult to detect through conventional pattern-matching approaches.
Featured Research: SEMA-GUARD
SEMA-GUARD: Semantic and Graph-Based Vulnerability Detection in Assembly Code
Dursunoglu & Sulkalar (2026) — arXiv Preprint
SEMA-GUARD investigates vulnerability detection in assembly language through semantic and graph-based representations of program behavior. The framework incorporates low-level program information, including control-flow relationships, stack operations, memory accesses, and data-flow patterns.
The research explores how graph-based machine learning can support vulnerability detection in compiled and low-level software systems.
Research Objectives
- Develop semantic representations of low-level program behavior.
- Investigate graph-based approaches to software vulnerability detection.
- Improve automated analysis of assembly and compiled software.
- Support secure software engineering through reproducible security evaluation.
3. Deepfake Detection, AI Impersonation & Digital Trust
Advances in generative artificial intelligence have made it increasingly difficult to distinguish authentic digital media from manipulated or synthetic content.
My research examines AI-generated media security, deepfake detection, audio-visual synchronization, multimedia authenticity, and digital trust.
A particular focus is identifying inconsistencies between speech, facial movements, and temporal synchronization that may indicate manipulated audio or video.
Featured Publication
Audio-Visual Synchronization Analysis for Deepfake Detection: A Comprehensive Review
Dursunoglu (2026)
International Journal of Computer Techniques
This work reviews synchronization-based approaches to deepfake detection and examines technical challenges involving temporal alignment, audiovisual consistency, detection reliability, and multimedia forensic evaluation.
I have also contributed expert commentary to RSA Conference (RSAC) on AI impersonation threats and synchronization-based approaches to strengthening digital trust.
Research Objectives
- Investigate methods for identifying manipulated audio-visual media.
- Examine synchronization inconsistencies as indicators of synthetic content.
- Improve understanding of AI-generated impersonation threats.
- Support trustworthy digital communication and media authenticity.
4. Trustworthy AI & Evidence-Grounded Systems
As AI-generated content becomes increasingly integrated into research, education, and professional decision-making, ensuring that generated information is supported by verifiable evidence is an important challenge.
My research interests in trustworthy AI include retrieval-augmented generation, evidence validation, citation verification, AI reliability, and the detection of unsupported generated claims.
Featured Research: CiteGuard-RAG
CiteGuard-RAG: A Validation-Centered AI System for Evidence-Grounded Question Answering
Barua, Hong, Dursunoglu, Rodgers & Fong (2026) — arXiv Preprint
CiteGuard-RAG explores a validation-centered approach to retrieval-augmented question answering. The framework combines information retrieval, citation-constrained generation, sentence-level evidence validation, and controlled answer regeneration to improve the grounding and reliability of AI-generated responses.
This research contributes to broader efforts to develop AI systems that provide transparent, verifiable, and evidence-supported information.
Research Objectives
- Improve the reliability of AI-generated responses.
- Investigate automated evidence and citation validation.
- Reduce unsupported claims in retrieval-augmented systems.
- Support trustworthy and verifiable AI applications.
5. AI-Driven Network Security & Automated Infrastructure
Artificial intelligence is increasingly used to support network management, configuration optimization, and infrastructure automation.
While these technologies offer opportunities for greater efficiency, incremental automated changes can introduce cumulative security risks that may be difficult to identify through periodic security reviews.
My emerging research investigates security drift in AI-driven network automation, with particular attention to configuration changes, access controls, segmentation, security policy consistency, and longitudinal security validation.
I am interested in methods that evaluate not only whether an individual configuration change satisfies immediate security requirements, but also whether a sequence of changes preserves the intended security posture over time.
Featured Professional Presentation
When AI Takes Control of the Network: Preventing Security Drift in Automated Infrastructure
BrightTALK Intelligent Network Summit — November 18, 2026
This upcoming presentation examines how AI-driven network operations can introduce cumulative security risks and discusses approaches to continuous validation, governance, and secure automation.
Research Objectives
- Investigate security drift caused by incremental automated network changes.
- Develop approaches to longitudinal security policy validation.
- Explore continuous monitoring and security-preserving remediation.
- Strengthen the reliability and governance of AI-managed infrastructure.
6. Human Activity Recognition & Intelligent Sensing
My research background also includes human activity recognition, wearable sensor systems, hybrid machine learning, and intelligent monitoring.
This work examines how sensor-generated data and computational models can be used to recognize human activities, including complex and composite activities in multi-user environments.
Selected Publications
Hybrid Classification for Complex and Composite Activity Recognition in Multi-User Environments
Dursunoglu et al. (2026)
IEEE Sensors Journal — Accepted for Publication
Human Activity Monitoring with Wearable Sensors and Hybrid Classifiers
Uslu, Dursunoglu, Altun & Baydere (2013)
International Journal of Computer Information Systems and Industrial Management
This research contributes to intelligent sensing, activity classification, and machine learning-based analysis of human behavior.
Cybersecurity Education & Student Research
Alongside my research, I contribute to cybersecurity education through experiential learning, curriculum development, and student mentorship at Western Michigan University.
My teaching and educational activities emphasize secure programming, network security, web application security, computer systems, and practical software engineering.
I have developed hands-on cybersecurity exercises, contributed to curriculum redesign, and mentored students working on software development and applied security projects.
These activities complement my research by connecting cybersecurity concepts with practical implementation and helping prepare students to address emerging security challenges.
Research Impact & Long-Term Goals
My research aims to advance the development of secure, reliable, and trustworthy digital technologies by addressing security challenges associated with artificial intelligence, software development, and increasingly automated systems.
My long-term research priorities include improving the security of AI-assisted software development, developing effective vulnerability detection techniques, strengthening the authenticity of AI-generated media, advancing evidence-grounded AI systems, and supporting security-preserving infrastructure automation.
Through applied research, interdisciplinary collaboration, reproducible experimentation, and engagement with the professional cybersecurity community, I seek to contribute practical methods that improve software security, digital trust, and the resilience of complex digital systems.
Research Resources & Academic Profiles
My publications, preprints, research software, and publicly released experimental materials are available through the following platforms:
Google Scholar — Publications and scholarly citations
ORCID — Researcher identification and academic record
GitHub — Research software, implementations, and repositories
ResearchGate — Research publications and academic networking
Publications Page — Peer-reviewed articles, conference papers, and preprints
Selected research artifacts are also archived through Zenodo, with repository links and persistent identifiers provided alongside the relevant publications.