Multimodal LLM Security, GPT-4V(ision), and LLM Prompt Injection Attacks

Original Image generated by author in DALL-E; a visual pun on the cybersec phrase “prompt injection”

Prompt injection attacks are a type of adversarial attack that exploit the vulnerability of large language models (LLMs) to malicious inputs. These attacks can manipulate the output of LLMs by injecting specially crafted prompts into the input, either dir…