Benchmark RadarAI BENCHMARK PROFILE
MLingualFC
MLingualFC evaluates jailbreak vulnerabilities in multilingual vision-language models using flowchart images encoding harmful instructions in five languages, measuring attack success rates.
- Released
- 2026-06-05
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
Safety alignment in multilingual VLMs is under-tested; this benchmark highlights gaps across languages, aiding safety evaluations, but lacks a standalone public comparison path.
Motivation
Vision-Language Models (VLMs) have demonstrated strong performance across multimodal tasks, yet their safety robustness remains an open challenge.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.