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