Entity · DiggingBeagle record
SecureVibeBench
Repository-level C/C++ secure-coding benchmark based on vulnerability-introducing scenarios.
Record description
SecureVibeBench reconstructs development scenarios from open-source vulnerability histories and asks coding agents to make multi-file changes. Its ACL 2026 paper combines functional testing with static and dynamic security checks, extending the comparison beyond small standalone code-generation tasks.
The benchmark helps test whether an implementation can both work and resist the evaluated security failures. It does not establish a production compromise rate or the security of unrelated languages and deployment platforms. SecureVibeBench is a separate project from the similarly named SecureVibe security-training research.