Author Type

Graduate Student

Date of Award

Summer 5-27-2026

Document Type

Dissertation

Publication Status

Version of Record

Submission Date

July 2026

Department

Computer and Electrical Engineering and Computer Science

College Granting Degree

College of Engineering and Computer Science

Department Granting Degree

Electrical Engineering and Computer Science

Degree Name

Doctor of Philosophy (PhD)

Thesis/Dissertation Advisor [Chair]

Mehrdad Nojoumian

Abstract

Securing interconnected software systems requires more than layering existing frameworks on top of each other. Most current Secure Software and System Development Lifecycle (S-SDLC) models treat security as a phase rather than a design condition, leaving real gaps in governance, access control, and automated enforcement that become critical failure points in Internet of Things (IoT) environments where devices are resource-constrained, long-lived, and frequently insecure by default.

This dissertation introduces the Automated Zero Trust Risk Management with DevSecOps Integration (AZTRM-D) methodology, a novel approach that unifies DevSecOps automation, the National Institute of Standards and Technology (NIST) Risk Management Framework (RMF), and Zero Trust (ZT) architecture into a single cohesive lifecycle with Artificial Intelligence (AI) as the orchestrating engine. AZTRM-D applies Zero Trust principles not only to users and devices but to its own AI components, addressing a blind spot present in every comparable framework evaluated. The methodology is operationalized through Cybectr Sentinel, an AI enforcement platform developed by Cybectr LLC, a United States Department of Defense contracting company, featuring six AI subsystems with formal mathematical specifications: Isolation Forest for behavioral anomaly detection, XGBoost with SHapley Additive exPlanations (SHAP) for vulnerability triage, Sentence Transformers for unknown asset inference, Proximal Policy Optimization (PPO) for AI-guided penetration testing, Retrieval-Augmented Generation (RAG) with a Large Language Model (LLM) for mitigation generation, and Diverse Counterfactual Explanations (DiCE) for adversarial robustness testing.

Empirical validation was conducted on physical NVIDIA Jetson Orin Nano hardware across three progressive hardening stages, with adversarial testing spanning hardware, radio frequency (RF), network, software, insider, privileged insider, and AI-assisted attack vectors. Testing was performed by three testers under a blind protocol with inter-rater agreement analysis (Gwet Agreement Coefficient 1 (AC1) = 0.888). All three testers are affiliated with Cybectr LLC; independent third-party laboratory validation is the highest-priority future-work deliverable. Factory-default devices were fully compromised to root level in under five minutes. Following full AZTRM-D hardening, zero successful breaches were recorded across any tested vector. The five-modality Continuous Integration/Continuous Delivery (CI/CD) scanning pipeline achieved a measured Vulnerability Detection Rate (VDR) of 96.8% (Wilson 95% Confidence Interval: [0.891, 0.991]). Cybectr Sentinel delivered 94.1% precision, 91.8% recall, a 3.1% aggregate False Positive Rate (FPR), and a 4.2-minute average Time to Initial Detection (TTID) within 12–18% Central Processing Unit (CPU) overhead on constrained edge hardware. The 93.7% adversarial detection rate is evaluated under a black-box threat model using DiCE-generated counterfactual inputs; gradient-based white-box evaluation is scoped to future work. A 14-capability comparative analysis against five established secure development frameworks found seven capabilities present in AZTRM-D that are absent from every evaluated alternative: AI-Assisted Code Analysis, Zero Trust Network Architecture, AI-Guided Penetration Testing, Unknown Asset Similarity Analysis, MITRE Detection, Denial, and Disruption Framework Empowering Network Defense (D3FEND) Mapping, MITRE Adversary Engagement, Deception, and Denial Framework (ENGAGE) Active Defense, and Post-Quantum Readiness Planning [69].

This dissertation also presents a post-quantum integration framework addressing the Harvest Now, Decrypt Later (HNDL) threat model, with concrete guidance for transitioning AZTRM-D’s cryptographic controls to NIST post-quantum standards including Module-Lattice-Based Key-Encapsulation Mechanism (ML-KEM) under Federal Information Processing Standards (FIPS) 203, Module-Lattice-Based Digital Signature Algorithm (ML-DSA) under FIPS 204, and Stateless Hash-Based Digital Signature Algorithm (SLH-DSA) under FIPS 205.

Download

Share

COinS