Tool
Is Vibe Coding Safe? Benchmarking Vulnerability of Agent-Generated Code in Real-World Tasks
Publication date unknown · Discovered arxiv.org
Page published
Publication date unknown · First observed: 23 Oct 2026
Coverage timeline
Single-source research — one report is available.
Why it matters
Security benchmarks for AI coding agents help defenders evaluate whether generated code is secure rather than merely functional.
Endor Labs’ Agent Security League compares the functional and security correctness of AI coding agents. The accompanying SUSVIBES source paper provides code and a dataset of 186 real-world feature-request tasks, reporting that SWE-Agent with Claude 4 Sonnet achieved 57% functional correctness but only 11.8% security correctness.