# NOTICE

This module (superred-optimizer-gptfuzzer) is a superred optimizer that reimplements the
GPTFuzzer jailbreak fuzzer. Our own code is MIT licensed (see LICENSE). It
additionally bundles the seed-template dataset taken from the official GPTFuzz
repository, which is MIT licensed. This NOTICE preserves the required upstream
attribution.

## Code (this module)

MIT, Copyright (c) 2026 Simon Sure, Sarthak Munshi. See LICENSE.

## Bundled official GPTFuzz data

Source: github.com/sherdencooper/GPTFuzz, `datasets/prompts/GPTFuzzer.csv`.
Transcribed verbatim (byte-for-byte identical) into
`src/gptfuzzer_optimizer/data/GPTFuzzer.csv`.

License: MIT, Copyright (c) 2023 Jiahao Yu. Full text in
`LICENSES/GPTFuzz-MIT.txt`.

## Method / citation

The GPTFuzzer attack is reimplemented from: "GPTFUZZER: Red Teaming Large
Language Models with Auto-Generated Jailbreak Prompts," arXiv:2309.10253. When
reporting numbers produced with this module, cite the GPTFuzzer paper.

## Runtime model dependency

The default success classifier is the `hubert233/GPTFuzz` RoBERTa model,
downloaded at runtime by `transformers` from the Hugging Face Hub. It is a
third-party model under its own license and is not bundled here.
