Attackers may infer or reconstruct original data by probing vector similarity APIs, especially with unrestricted access. This enables model inversion attacks, membership inference, and unauthorized data leakage from stored embeddings.
Database / Vector / Threats / DEV
Embedding Extraction and Model Inversion
CCC.Vector.TH01
Related Capabilities
| ID | Title | Description |
|---|---|---|
| CCC.Vector.CP01 | Embedding Storage | Supports storage of high-dimensional vector embeddings derived from raw input data such as text, images, or audio. |
| CCC.Vector.CP03 | Similarity Search | Enables nearest-neighbor queries using a query embedding to return the most similar vectors from the store. |
| CCC.Vector.CP06 | Real-Time Querying | Provides low-latency response to vector similarity queries suitable for interactive applications. |
External Mappings
| Framework | ID | Remarks |
|---|---|---|
| FINOS-AIGF | AIR-SEC-002 | Information Leaked to Vector Store |