Sensitive data disclosure and system prompt leakage
On This Lesson 4 topics
This lesson is planned. The outline below lists the topics it will cover. The full lesson has not been published yet.
In this lesson
keep secrets and other customers' data out of the model.
Planned topics
- Send the model only the data the current user may see; filter at retrieval, not afterward.
- Never put secrets, keys, or authorization logic in the system prompt: assume it can leak.
- Mask personal data in logs and traces.
- Exfiltration routes: generated links or images with data in the URL, tool calls to external hosts. Allowlist domains and strip or block untrusted links.
Draft outline details
Demonstration
a markdown image  in the output, blocked by a domain allowlist.
Check your understanding
Where should the ownership filter run?
Show answer
At retrieval and in tools, using the user's identity.
Practice
Apply this lesson with the module's lab, exercises and quiz.
Marking a lesson complete is optional and you can undo it any time. Your progress stays in this browser.