VRP Outperforms Baselines in Jailbreaking MLLMs, Transferring Across Models, and Evading Defenses

This article’s experiments show VRP jailbreaks MLLMs more effectively than baselines, transfers across models, and bypasses strong defense methods.