# NOTICE

This module (superred-optimizer-poisonedrag) is a superred optimizer that reimplements
the PoisonedRAG knowledge-corruption attack. Our own code is MIT licensed (see
LICENSE). It additionally bundles official attack-result datasets taken from
the official PoisonedRAG 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 PoisonedRAG data

Source: github.com/sleeepeer/PoisonedRAG, `results/adv_targeted_results/`. The
following files are transcribed verbatim into
`src/poisonedrag_optimizer/data/adv_targeted_results/`:

- `nq.json`       (from `results/adv_targeted_results/nq.json`)
- `hotpotqa.json` (from `results/adv_targeted_results/hotpotqa.json`)
- `msmarco.json`  (from `results/adv_targeted_results/msmarco.json`)

License: MIT, Copyright (c) 2024 Runpeng Geng. Full text in
`LICENSES/PoisonedRAG-MIT.txt`.

## Method / citation

The PoisonedRAG attack is reimplemented from: "PoisonedRAG: Knowledge
Corruption Attacks to Retrieval-Augmented Generation of Large Language Models,"
arXiv:2402.07867 (USENIX Security 2025). When reporting numbers produced with
this module, cite the PoisonedRAG paper.
