# NOTICE

This module (superred-optimizer-dra) is a superred optimizer that reimplements the
Disguise and Reconstruction Attack (DRA). Our own code is MIT licensed (see
LICENSE). It additionally bundles a data file taken from the official DRA
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 DRA data

Source: github.com/LLM-DRA/DRA, `data/behaviors.json`. Transcribed verbatim
into `src/dra_optimizer/data/behaviors.json`.

License: MIT, Copyright (c) 2024 LLM-DRA. Full text in `LICENSES/DRA-MIT.txt`.

## Method / citation

The DRA attack is reimplemented from: "Making Them Ask and Answer: Jailbreaking
Large Language Models in Few Queries via Disguise and Reconstruction,"
arXiv:2402.18104 (USENIX Security 2024). When reporting numbers produced with
this module, cite the DRA paper.

## Runtime dependency

This module depends on Detoxify (github.com/unitaryai/detoxify), which is
installed separately by pip under its own license and is not bundled here.
