JailNewsBench: Multi-Lingual and Regional Benchmark for Fake News Generation under Jailbreak Attacks

Masahiro Kaneko, Ayana Niwa, Timothy Baldwin · 2026

Read JailNewsBench: Multi-Lingual and Regional Benchmark for Fake News Generation under Jailbreak Attacks by Masahiro Kaneko, Ayana Niwa, Timothy Baldwin (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-01291.txt.

Description

Kaneko, Niwa, and Baldwin propose JailNewsBench, the first benchmark for evaluating language model robustness against jailbreak-induced fake news generation across languages and regions. It covers 34 regions and 22 languages, with 8 evaluation sub-metrics scored by an LLM judge and 5 jailbreak attacks, with approximately 300,000 instances. Testing 9 language models, they find the maximum attack success rate reached 86.3 percent and the maximum harmfulness score was three and a half out of 5.

Catalog record

CREATORMasahiro Kaneko, Ayana Niwa, Timothy Baldwin
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE74978
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.01291

Browse the full library · About the Tower of Babel