A system-level approach to prompt injection: separating instruction and data channels in LLM agents [P]
8/10This article proposes a system-level method to prevent prompt injection in large language model agents by separating instruction and data channels, addressing a major failure mode in tool-using LLM systems, especially in external data interactions.
