ConcurBugBench

Concurrency bug discovery and repair

Posted: July 01, 2026

Tags: Benchmark, Agents, Systems, Security

ConcurBugBench tests agent capabilities to locate, reproduce, and fix authentic concurrency bugs where failures stem from specific thread interactions, timing, and shared state dynamics. Each bug includes a validation mechanism for both reproduction and repair.

~200 real bugs sourced from production codebases
Languages: Go, Rust, C, C++
Bug-specific oracles for deterministic validation