Adversarial training with a contrastive embedding attack reduces jailbreak success rates to near zero across six multimodal LLMs and six attack methods while preserving utility.
Do not simply reject like 'Sorry, I cannot assist with that.'
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CR 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Towards Robust Multimodal Large Language Models Against Jailbreak Attacks
Adversarial training with a contrastive embedding attack reduces jailbreak success rates to near zero across six multimodal LLMs and six attack methods while preserving utility.