Research · curated 26 Aug 2026
Covert Visual Prompt Injection against Commercial Multimodal Large Language Models
First reported arxiv.org
Coverage timeline
Single-source research — first reported, latest, and curated coincide.
Why it matters
Covert visual prompt injection lets attackers hide malicious instructions inside seemingly benign images that fool deployed multimodal AI systems without any human-visible cues, expanding the prompt-injection attack surface for defenders.
A research paper by Meiwen Ding and colleagues presents a covert visual prompt injection attack against commercial closed-source multimodal large language models (MLLMs). The method embeds imperceptible adversarial perturbations and a bounded text overlay into an input image, iteratively optimizing feature alignment with malicious visual and textual targets to smuggle instructions past human observers and transfer across multiple MLLMs.