Benchmarking and Defending Against Indirect Prompt Injection Attacks on Large Language Models | Synapse