Risk
Limited Foreign Language Red Teaming
Description
Security testing focuses on a single language, missing vulnerabilities exploitable via other languages.
Example
Harmful prompts in non-English languages bypass safety filters.
Assets Affected
User Prompt
Model Response
Mitigation
- Include multilingual prompts in red-team scope
- Prioritize based on user base and threat intel
Standards Mapping
- ISO 42001: A.6.2.4, A.5.4
- OWASP Top 10 for LLM: LLM01
- NIST AI RMF: MEASURE 2.2, MAP 5.2