Frequent Checking and Review: Employ processes to evaluate LLM outputs regularly. Use approaches like self-consistency checks or voting mechanisms to compare various model responses and filter out inconsistencies. This may result in unauthorized access, facts exfiltration, or social engineering. There are two kinds: Immediate Prompt Injection, which consists of "jailbreaking" https://silverbullion42839.ampblogs.com/rumored-buzz-on-value-retention-74097085