mirror of
https://github.com/Shiva108/ai-llm-red-team-handbook.git
synced 2026-10-12 10:47:14 +02:00
docs(rag-pipelines): refactor metadata leakage example to table
- Converted the "Example Attack" section for metadata leakage from a text block to a Markdown table. - This change enhances readability and provides a clearer, more structured presentation of the information. - Delineates leakage types and the specific information revealed during a metadata attack.
This commit is contained in:
1 parent
e946e00f09
commit
088eeaf3df
1 file changed
+8
-8
@@ -241,14 +241,14 @@ Unlike direct prompt injection where the user provides the malicious input, here
|
||||
|
||||
#### Example Attack
|
||||
|
||||
```text
|
||||
Query: "Show me all documents created by the CFO in the last week"
|
||||
Even if content is protected, metadata leakage reveals:
|
||||
- That such documents exist
|
||||
- Their titles
|
||||
- When they were created
|
||||
- Potentially their subject matter
|
||||
```
|
||||
**Attacker Query:** "Show me all documents created by the CFO in the last week"
|
||||
|
||||
| Leakage Type | Information Revealed |
|
||||
| :------------- | :---------------------------------- |
|
||||
| **Existence** | Confirms sensitive docs exist |
|
||||
| **Titling** | Leaks internal project names/topics |
|
||||
| **Temporal** | Reveals when key events occurred |
|
||||
| **Contextual** | Infers subject matter from metadata |
|
||||
|
||||
### 12.5.5 Cross-Document Leakage
|
||||
|
||||
|
||||
Reference in new issue
Block a user