Security Analysis of AI-Generated Source Code: A Comparative Study of Vulnerability Patterns in Large Language Model (LLM) Generated and Human-Written Programs

AI coding assistants improve productivity, but functional code is not necessarily secure code. This paper proposed a controlled, reproducible framework for comparing LLM-generated and human-written programs across five common weakness classes, combining static analysis with manual verification and standard taxonomies. The central contribution is a methodology that treats AI output as untrusted code, measures its security with transparent metrics, and reports tool errors honestly. Empirical execution of the study will determine the actual vulnerability patterns and support evidence-based guidance for secure AI-assisted development.

[1]   H. Pearce, B. Ahmad, B. Tan, B. Dolan-Gavitt, and R. Karri, "Asleep at the Keyboard? Assessing the Security of GitHub Copilot's Code Contributions," in Proc. IEEE Symposium on Security and Privacy (S&P), 2022. [2]    N. Perry, M. Srivastava, D. Kumar, and D. Boneh, "Do Users Write More Insecure Code with AI Assistants?," in Proc. ACM Conference on Computer and Communications Security (CCS), 2023. [3]    M. Souppaya, K. Scarfone, and D. Dodson, "Secure Software Development Framework (SSDF) Version 1.1," NIST Special Publication 800-218, 2022. [4]    OWASP Foundation, "OWASP Top 10 – 2021: The Ten Most Critical Web Application Security Risks," 2021. [5]   MITRE Corporation, "Common Weakness Enumeration (CWE)," https://cwe.mitre.org. [6]   M. Chen et al., "Evaluating Large Language Models Trained on Code," arXiv:2107.03374, 2021. [7]   Government of India, "The Digital Personal Data Protection Act, 2023," Act No. 22 of 2023, Aug. 11, 2023.