System prompt leak
A system prompt leak is a model revealing its own hidden instructions when a user asks for them directly or indirectly.
System prompts should be treated as visible by default. Instructions not to reveal them are unreliable, so anything genuinely secret — keys, internal URLs, unpublished policy — must not be in the prompt at all.
Related terms
- System prompt
- A system prompt is a persistent instruction that sets a model’s role, rules, and boundaries for an entire conversation, separate from the individual messages a user sends.
- Jailbreak
- A jailbreak is a prompt crafted to make a model bypass its own safety training and produce output it would normally refuse.
More on safety
See the full glossary, read the guides, or put it into practice in the prompt builder.