Chemotactic Test Case Recombination for Large-Scale Fuzzing

We present a bio-inspired method for large-scale fuzzing to detect vulner-abilities in binary executables. In our approach we deploy small groups of feedback-driven explorers that guide colonies of high throughput fuzzers to promising regions in input space. We achieve this by applying the biological concept of chemotaxis: The explorer fuzzers mark test case regions that drive the target binary to previously undiscovered execution paths with an attrac-tant. This allows us to construct a force of attraction that draws the trailing fuzzers to high-quality test cases. By introducing hierarchies of explorers we construct a colony of fuzzers that is divided into multiple subgroups. Each subgroup is guiding a trailing group and simultaneously drawn itself by the traces of their respective explorers. We implement a prototype and evaluate our presented algorithm to show the feasibility of our approach.