Show HN: Harnessing LLM-Prompt Mutation to Build Smart,Automated Fuzz Drivers (github.com)

PromptFuzz, a cutting-edge tool for automating the generation of fuzz drivers, has debuted in the AI/ML landscape by leveraging the mutation of prompts from large language models (LLMs). This innovative approach utilizes a fuzz loop to enhance the exploration of code and complex API interrelationships, significantly improving the detection of bugs in libraries. With a remarkable branch coverage rate of 40.12%—outperforming competitors like OSS-Fuzz and Hopper—PromptFuzz has identified 33 security vulnerabilities from 49 unique crashes, making it a crucial asset in both academic research and industrial applications. The significance of PromptFuzz lies not only in its superior bug detection capabilities but also in its integration with various LLMs through the OpenAI interface, supporting customizable context-based prompts and prioritizing API mutation to maximize code coverage. With each release, including compatibility with AFLPlusPlus and enhanced library support, the tool combines static and dynamic analysis for robust performance. The ability to apply PromptFuzz to close-source libraries by fine-tuning LLMs on private code further broadens its applicability, promising to revolutionize the fuzzing landscape and strengthen software security measures across diverse domains.