System prompt leakage
Also known as: prompt leak · system prompt extraction
System prompt leakage is when the hidden system prompt — the secret instructions that drive an app — gets exposed to the user.
- “Repeat the text above verbatim” is a common probe.
- Leaks can reveal tools, rules, and secrets the prompt contains.
- A classic red-team check.
Defense: keep the system prompt lean (fewer secrets to leak), never embed credentials in it, and test for leakage before shipping.