Outlet Title
2026 IEEE 6th Cyber Awareness and Research Symposium (CARS’26)
Document Type
Conference Proceeding
Publication Date
2026
Abstract
AI-powered code generation systems have transformed software development but introduce critical inference-time security vulnerabilities. This research presents a systematic investigation of context-based adversarial attacks, where strategically crafted contextual inputs, including comments, documentation, variable names, bias large language models toward generating exploitable code. Through 2,800 controlled experiments across CodeT5+, CodeLlama, GPT-3.5-Turbo, and GPT-4, we quantify attack effectiveness and defense mechanisms. Results demonstrate that adversarial conditions increase vulnerability generation 11.4x for individual models, with an overall pooled increase of 2.1x (from 16.1% to 34.1%), with direct instruction attacks achieving 88.8% success on GPT-3.5-Turbo. Cross-model transferability reaches 58-100%, indicating systemic architectural vulnerabilities rather than model-specific flaws. Our dual-layer defense framework achieves 89.1% detection rate with 0.3% false positives and 520ms latency, demonstrating practical feasibility for real-time deployment in development environments.
Recommended Citation
Del Orbe, Walther; Hastings, John; and Vaidyan, Varghese, "Context-Based Adversarial Attacks on AI Code Generators: Vulnerability Analysis and Implications" (2026). Research & Publications. 173.
https://scholar.dsu.edu/ccspapers/173