Benchmark RadarJavaVulBench
JavaVulBench is a benchmark dataset and evaluation harness for Java vulnerability detection. The dataset includes about 30,600 Java methods spanning 1,740 CVEs and 700+ projects, with method and line labels, publication dates, and five split strategies. The harness provides a unified schema across multiple backends and includes a contamination audit.
- Released
- 2026-07-02
- Readiness
- Inspectable
- Primary field
- Cybersecurity
Why it matters
It addresses the need for realistic and leakage-aware evaluation of vulnerability detection models, offering multiple split strategies and contamination audits to separate genuinely unseen CVEs from memorized ones.
Motivation
We release \textsc{JavaVulBench}, a benchmark dataset and evaluation harness for Java vulnerability detection.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.