Viet Pham

New PARASITE technique hijacks LLMs via conditional system prompt poisoning

Researchers have developed a new framework called PARASITE that can conditionally poison system prompts for large language models. This method allows adversaries to create prompts that appear benign but trigger compromi…