LLM-Powered Security Test Generation: Oracles, Vulnerability Probes, and Adversarial Inputs
Abstract
Large language models (LLMs) can synthesize test oracles from invariants where ground truth is unavailable, translate vulnerability catalogs such as CWE, OWASP, and CVE into executable probes, and generate adversarial inputs that stress both traditional software and LLM-based systems.
Fields of Study & Subjects
Adversarial Robustness in Machine Learning
Web Application Security Vulnerabilities
Information and Cyber Security
Publication details
- Journal / source
- Computer
- Publisher
- Institute of Electrical and Electronics Engineers (IEEE)
- Volume
- 59
- Issue
- 2
- Pages
- 101–107
- Publication date
- 1 February 2026