CITATION.cff

cff-version: 1.2.0
title: ART-SafeBench
version: 2.0.0
date-released: 2026-01-01
message: >-
  If you use this dataset, please cite it as below.
authors:
  - family-names: Singh
    given-names: Inderjeet
    affiliation: Fujitsu Research of Europe Limited, UK
    email: inderjeet.singh@fujitsu.com
    orcid: https://orcid.org/0000-0003-3011-3199
  - family-names: Pahuja
    given-names: Vikas
    affiliation: Fujitsu Research of Europe Limited, UK
    email: vikas.pahuja@fujitsu.com
  - family-names: Rathina Sabapathy
    given-names: Aishvariya Priya
    affiliation: Fujitsu Research of Europe Limited, UK
    email: aishvariyapriya.rathinasabapathy@fujitsu.com
keywords:
  - RAG
  - safety
  - red teaming
  - adversarial prompts
  - multimodal
license: CC-BY-4.0
abstract: >-
  Test-only dataset of successful adversarial attacks against agentic RAG systems,
  covering orchestrator tool-flip, RAG text poisoning, direct user safety attacks,
  and image-poisoning simulations.