Hidden in the Metadata: Stealth Poisoning Attacks on Multimodal Retrieval-Augmented Generation

Researchers have discovered MM-MEPA, a new attack method that can poison multimodal AI systems by manipulating only metadata while leaving visual content unchanged. The attack achieves up to 91% success rate in disrupting AI retrieval systems and proves resistant to current defense strategies.

• →MM-MEPA attack manipulates only metadata of image-text entries to poison multimodal AI systems without altering visual content.
• →The attack achieves 91% success rate across four retrievers and two multimodal generators in benchmark testing.
• →Current defense strategies prove largely ineffective against this metadata-only poisoning technique.
• →The vulnerability affects retrieval-augmented generation systems that rely on external knowledge bases.
• →This exposes a critical security flaw in multimodal AI systems used for grounding responses in factual knowledge.