New AI attack hides data-theft prompts in downscaled images

An AI-based attack technique enables data theft by embedding malicious prompts within images. These prompts are revealed only after the images are downscaled by AI systems, allowing attackers to inject commands into LLMs without user awareness.