How LLM Applications Leak Data Through Prompt Responses
System prompts leak easily because models can't distinguish help requests from extraction attempts.
Marco Quist
Section
Data Exfiltration in LLM Applications
1 story in Data Exfiltration in LLM Applications.