New Framework Optimizes Bug Detection by Intelligently Managing Multiple Fuzzing Tools

Researchers have developed Legion, an ensemble fuzzing framework designed to more efficiently combine the strengths of multiple bug-detection tools while minimizing wasted computational resources.

The Challenge with Current Approaches#

Fuzzing—a technique that feeds random or crafted inputs to software to expose bugs and security flaws—has become central to vulnerability discovery. While combining multiple fuzzing strategies into an ensemble can improve results, existing approaches struggle with two key problems: they allocate resources inefficiently across different fuzzers, and they lack nuanced ways to evaluate which fuzzing techniques are actually working.

How Legion Works#

The framework introduces a dynamic resource scheduling algorithm inspired by the upper confidence bound method, which learns which base fuzzers are performing well and redirects computational power away from underperformers. This approach cuts down on wasted effort without sacrificing coverage.

Complementing this is a multidimensional seed evaluation strategy that assesses fuzzing progress using multiple metrics rather than relying on a single measurement. This finer-grained assessment allows the system to better understand what each fuzzer contributes and make smarter scheduling decisions.

Real-World Validation#

When tested on Google's fuzzer-test-suite and various open-source projects, Legion detected 20 previously unknown or officially recorded bugs, including five that had never been disclosed before and three that earned CVE (Common Vulnerabilities and Exposures) designations. The results demonstrate that Legion outperforms both existing standalone fuzzers and competing ensemble approaches.

The research, submitted to arXiv on July 30, 2025, suggests that intelligent resource management and comprehensive performance measurement can meaningfully improve the speed and effectiveness of automated vulnerability discovery.