OpenAI's new GPT-4V release supports image uploads — creating a whole new attack vector making large language models (LLMs) vulnerable to multimodal injection image attacks. Attackers can embed ...