Release v1.0.0

Official Release v1.0.0

About bug fixing

This official release resolved a number of panic bugs (#39, #38, #41, #40, #30, #26) that were reported by users last year. We also made a lot of attempts to avoid PromptFuzz panic when it encounters corner cases in LLM-generated harnesses (79d4bb8, 212689b, b3d0931, a3565bd, 58c2d79, 368cf80, 07ecf25, 7519850, f7773cd, 7574ce0, 42f8e0d).

About network isolation

Since testing network libraries could be dangerous, such as testing the DNS parsing library (c-ares), even in the Docker environment. We cannot disable the entire network for Docker containers as we need to send LLM requests, so we adopted bwrap to isolate the network effect in the library testing procedures (e807559, df5a064, 16cde61).

Support new libraries

Some users asked for the support of more popular libraries, we supported these new ones: expat, ffmpeg, freetype, liblouis, libpsl, libxml2, libyaml, opencv, openssl, zstd. To support these libraries, each one costs us nearly 8 hours. This is truely a hard work as we need to handle the library building and various fuzzing instrumentation properly. Thanks to those contributors who provided their build scripts.

Support AFLPlusPlus

We provided scripts to support AFLPlusPlus in fuzzing PromptFuzz-generated drivers. See AFLPlusPlus how to use it.

V0.0.1

1. Add new libraries: ffmpeg, liblouis, OpenSSL
2. Fix build issues in existing libraries
3. Protect the network in fuzzing by bubblewrap (Fuzzing c-ares will cause DoS to networks)
4. Fix some existing issues that are triggered often.

V0.0.0

Official release version of PromptFuzz's Paper

We are pleased to announce the official release of our paper, PromptFuzz. Given the rapid advancements in AI technologies, some aspects of PromptFuzz's implementation may become outdated. To address this, we are committed to continuously updating PromptFuzz with the latest AI technologies, such as advanced LLM models, and introducing new features.

To facilitate replication and comparison studies, we are releasing the version of PromptFuzz used in our evaluation. This will enable researchers to effectively build upon or compare their work against our initial findings.

If there were some bugs in the code of this version, we will fix them in later versions.