Large Vision Language Models (LVLMs) are susceptible to a sophisticated attack known as the bi-modal adversarial prompt (BAP). This attack exploits the integration of textual and visual prompts to circumvent established safety protocols, thereby eliciting harmful outputs from models engineered to withstand single-modality threats. The attack initiates by embedding a query-agnostic adversarial perturbation within the visual prompt, increasing the likelihood of the model's affirmative response irrespective of the accompanying text. Subsequently, a language model iteratively refines the textual prompt to fulfill the intended malicious objective. Mitigation steps: **For AI Developers:** * Implement advanced safety mechanisms to detect and mitigate attacks targeting both visual and textual components of prompts. * Apply stringent input validation and filtering processes to identify and reject potentially harmful prompts. **For Model Trainers/Fine-tuners:** * Enhance model robustness against adversarial attacks by integrating methods that address combined textual and visual manipulations. * Continuously update and refine safety models to counteract evolving attack strategies, incorporating adversarial training techniques to defend against bi-modal attacks.