Differences in Safety Risks Across Languages in Health-Relevant Queries: Vulnerability Analysis of Large Language Model Responses

Saubhagya Joshi, Monjil A Mehta, Melissa Mendoza, Yonaira M Rivera, Vivek K Singh · 2026

Read Differences in Safety Risks Across Languages in Health-Relevant Queries: Vulnerability Analysis of Large Language Model Responses by Saubhagya Joshi, Monjil A Mehta, Melissa Mendoza, Yonaira M Rivera, Vivek K Singh (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/pmc-13211943.txt.

Description

Chatbot safety filters work unevenly across languages, and a 2026 vulnerability analysis led by Saubhagya Joshi measures the gap with disguised harmful prompts. The study attacks a health-oriented language model in English, Spanish, and transliterated Hindi, using emoji and letter-permutation ciphers across self-harm, violence, and drug abuse categories.

One thousand prompts per language became six thousand model interactions for human coders to judge. Hindi prompts broke through most often under both disguise styles, with Spanish and English resisting better in the comparisons. Violence-related prompts succeeded more than the other harm types overall, and the authors urge stronger multilingual moderation before such tools guide health decisions.

Catalog record

CREATORSaubhagya Joshi, Monjil A Mehta, Melissa Mendoza, Yonaira M Rivera, Vivek K Singh
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE40 KB
VERSION1.0
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC13211943

Browse the full library · About the Tower of Babel