arXiv:2609.22949v1 Announce Type: new Abstract: Existing prompt injection research focuses on single-model chatbot scenarios, where an attacker manipulates one LLM through crafted input.

Read the full article at arXiv cs.CR →