paper

Exploring Adversarial Robustness and Safety Alignment in Multilingual Multi-Modal Large Language Models

paperactiveprovisionalexploring-adversarial-robustness-and-safety-alignment-in-multilingual-multi-modal-large-language-models-7633268e·1 events·first seen 13d ago

Aliases: Exploring Adversarial Robustness and Safety Alignment in Multilingual Multi-Modal Large Language Models

Co-occurring entities

Qwen3-4B

More like this (12)

The Shibboleth Effect: Auditing the Cross-Lingual Distributional Skew of Large Language Models Multimodal Large Language Models Automated reproducibility assessments in the social and behavioral sciences using large language models Contrastive-Difference CKA Reveals Concept-Specific Structural Alignment Across Language Model Architectures Language Model Safety Monitor multi-turn language models adversarial robustness Adaptive Multi-Resolution Procedural Knowledge Compression for Large Language Models Civil Court Simulation with Large Language Models Reinforcement Learning for Language Models The Neutral Mask: How RLHF Provides Shallow Alignment while Leaving Partisan Structure Intact in a Large Language Model Multi-Faceted Interactivity Alignment in Full-Duplex Speech Models

Recent events (1)

6arXiv · cs.CL·13d ago·source ↗

Adversarial robustness and safety alignment in multilingual multimodal LLMs: cross-lingual vulnerability and 'safety-by-failure'

A systematic study evaluates adversarial robustness and safety alignment of multimodal LLMs across 12 languages, finding that adversarial images optimized in one language transfer to others (cross-lingual transferability). The paper introduces the concept of 'safety-by-failure': low-resource languages appear safer not due to genuine alignment but because models fail to comprehend harmful instructions in those languages. Models like Qwen3-VL that integrate multilingual capability throughout training (rather than only at instruction tuning) show genuine cross-lingual safety with active refusal. The findings challenge the assumption that low-resource language safety metrics reflect real alignment.

Evaluation and Benchmarking AI Safety Research Qwen3-4B Exploring Adversarial Robustness and Safety Alignment in Multilingual Multi-Modal Large Language Models +1 more