🖼️2310Read later0Read MorePrompt Injection as Role ConfusionHacker News·4 months ago#l5hU3EhN#role-confusion#role#user#roles#reasoning+4 more🧰Tag tools✨Add tagLLMs can't tell who's speaking. We show they identify roles by writing style, not tags, and exploit this with CoT Forgery, injecting fake reasoning that models mistake for their own thoughts.15s
