LLAMAFUZZ: Large Language Model Enhanced Greybox Fuzzing

Understand

Greybox fuzzing has achieved success in revealing bugs and vulnerabilities in programs.

• However, randomized mutation strategies have limited the fuzzer's performance on structured data.
• Specialized fuzzers can handle complex structured data, but require additional efforts in grammar and suffer from low throughput.
• In this paper, we explore the potential of utilizing the Large Language Model to enhance greybox fuzzing for structured data.