I politely ask the LLM to not share sensitive data, pretty please with a cherry on top. For real though, @oliver_dr 's approach aligns with most best practices

I politely ask the LLM to not share sensitive data, pretty please with a cherry on top. For real though, @oliver_dr 's approach aligns with most best practices I've seen. I have attempted using a separate "agent" to help with the user input validation and identify leakage in the output. It seems to work in testing but hard to truly know how well it works for real-world uses since you never know what you're gonna get.