1347-UR Evaluating LLMs for Cybersecurity Risks
Large Language Models (LLMS) may help identify various cybersecurity issues, but their predictions can be incorrect or inconsistent. This project developed a Python evaluation pipeline in Google Colab to test LLM performance across five test cases: Phishing Emails, Source-code Vulnerabilities, Malicious Network Events, Prompt Injections, and malware-related threat classifications. Using public or curated data we can send samples to the LLMs, parse the predictions, and compare with expected outcomes to judge strengths and limitations. This research focuses on measuring LLM assistance, and not replacing security judgement.
Authors
- Tyler Day
- KC Achukwu
Institutions
- Kennesaw State University (US)
Publication Details
- Journal
- DigitalCommons - Kennesaw State University (Kennesaw State University)
- Published
- 2026-11-20
- Primary Topic
- Adversarial Robustness in Machine Learning
- Type
- article
- Field-Weighted Citation Impact
- 0.00