pith. sign in

Jailbreak large vision-language models through multi-modal linkage

1 Pith paper cite this work. Polarity classification is still indexing.

1 Pith paper citing it

citation-role summary

background 1

citation-polarity summary

fields

cs.LG 1

years

2026 1

verdicts

UNVERDICTED 1

roles

background 1

polarities

background 1

representative citing papers

Dictionary-Aligned Concept Control for Safeguarding Multimodal LLMs

cs.LG · 2026-04-10 · unverdicted · novelty 6.0

DACO curates a 15,000-concept dictionary from 400K image-caption pairs and uses it to initialize an SAE that enables granular, concept-specific steering of MLLM activations, raising safety scores on MM-SafetyBench and JailBreakV while preserving general capabilities.

citing papers explorer

Showing 1 of 1 citing paper.

  • Dictionary-Aligned Concept Control for Safeguarding Multimodal LLMs cs.LG · 2026-04-10 · unverdicted · none · ref 109

    DACO curates a 15,000-concept dictionary from 400K image-caption pairs and uses it to initialize an SAE that enables granular, concept-specific steering of MLLM activations, raising safety scores on MM-SafetyBench and JailBreakV while preserving general capabilities.