Multimodal AI models pose increased risks for abuse and harmful content, study finds
Enkrypt AI's study reveals that new jailbreak techniques exploit multimodal models' media processing, bypassing filters and producing harmful outputs without visible prompts.
