Dictionary entry
LLM-driven fuzzing
also LLM-assisted fuzzing or LLM-guided fuzzing or LLM fuzzing
Definition of LLM-driven fuzzing
-
: a software-testing technique that uses a large language model to create or improve fuzz targets, seeds, mutations, test harnesses, protocols, or feedback so that automated testing can explore more program behaviors and find bugs
- The security team used LLM-driven fuzzing to generate structured inputs that passed the parser's initial checks and reached deeper code paths.
- LLM-assisted fuzzing produced a test harness, but conventional instrumentation still determined whether the new inputs increased code coverage.
Origin & history
Researchers began combining large language models with fuzz testing in the early 2020s. Systems such as ChatAFL, presented at NDSS 2024, used an LLM to understand protocol message formats and guide mutations, while later work expanded the approach to harness generation, seed creation, program analysis, and automated security-testing agents.
Test yourself
Which of these is the meaning of LLM-driven fuzzing?
Cite this entry
"LLM-driven fuzzing." AI Dictionary, Dadgogo, https://dadgogo.com/dictionary/llm-driven-fuzzing/. Accessed 9 Oct. 2026.