SNIPTEST automates vulnerability validation with 54.6% accuracy

Researchers from [institution names omitted for brevity] have developed **SNIPTEST**, an automated framework designed to validate static-analysis warnings by generating and fuzzing compiled code slices around potential vulnerabilities. Unlike traditional approaches that require manual inspection or full-program analysis—which can take days—SNIPTEST uses a layer-by-layer slicing strategy to incrementally expand execution context around target locations. This method provides evidence about how warnings behave under progressively refined conditions, improving both accuracy and efficiency.

In evaluations across 194 cases (97 confirmed vulnerabilities and 97 false alarms) from three real-world projects, SNIPTEST demonstrated strong performance: it produced possible true positive evidence for 54.6% of confirmed vulnerabilities by consistently triggering bug oracles across all slice levels, and exploited vulnerabilities in 40.2% of cases. For false alarms, it correctly identified 55.6% as likely false positives by reaching warnings without triggering bug oracles. Notably, SNIPTEST also contributed to the discovery of CVE-2025-11964, highlighting its practical relevance in real-world security research.