Evaluating the Instruction-Following Robustness of Large Language Models to Prompt Injection

Open in new window