The chatbot that shared its instructions
A user posted a screenshot of your support bot's hidden instructions, including the staff-only discount code and an .
This answers: Can users see my chatbot's system prompt? No coding needed. 3 quick decisions.
- Keys and secrets
- Supervising your AI
Checking your access…
Can users see my chatbot's system prompt?
Assume yes. A system prompt is text in the same conversation as the user's messages, and models can be talked into repeating it. The prompt text itself is rarely the danger; secrets and rules inside it are. Keep in server and enforce permissions in code, so a leaked prompt reveals nothing useful.
Breaks at Your server: Secrets placed inside the prompt
Why AI tools write this bug
Putting everything the bot needs into its instructions is the quickest way to make it work, so keys and internal rules often end up there.
How to fix it
- Remove keys, codes and internal URLs from prompts.
- Revoke anything that was in a prompt.
- Enforce permissions in tools and server code.
- Treat the prompt as public text.
Check your own app
Paste this into Lovable, Cursor or Claude Code:
Without changing anything, show me every system prompt and instruction text this project sends to an AI model. Flag anything a user shouldn't see: keys, internal URLs, discount codes, staff rules, other customers' data. For each rule in a prompt, tell me whether code also enforces it.
Common questions
- Can I stop the model revealing its prompt?
- You can make it less likely, but not impossible. Design as if the prompt is public.