| CCC.Vector.TH01 | Source Data is Reconstructed from Stored Embeddings | Embeddings stored by the service or returned by similarity queries may be probed repeatedly by a user with query access, or exported in bulk, and inverted with a reconstruction model. The text, images, or personal data from which the embeddings were derived are recovered, and the presence of specific records in the index is inferred. Confidentiality of the source data is lost even though the raw data was never stored in the service. | 1 | 1 | 1 |
| CCC.Vector.TH02 | Index is Poisoned by Crafted Embeddings or Content | Vectors or source content submitted through ingestion endpoints, pulled from connected data sources, or vectorized by an integrated embedding model may be crafted so that they rank highly for targeted queries. Search results are steered toward chosen records or degraded for legitimate ones, and downstream applications consume the manipulated results. Integrity of the index and of every system that relies on it is compromised. | 1 | 1 | 1 |
| CCC.Vector.TH03 | Query Results Include Records the Requester is Not Authorized to Read | Document-level access controls may be enforced on one query path but not applied to vector similarity, hybrid, or metadata-filtered queries, or metadata filters may be used to confirm attribute values of records the requester cannot retrieve. Records, metadata fields, or correlations across vector fields derived from different modalities are returned or inferred. Confidentiality of the stored records is lost. | 1 | 1 | 1 |
| CCC.Vector.TH04 | Queries are Served from a Replaced or Stale Index | The index behind a query endpoint or alias may be swapped, redeployed, or re-pointed by an identity with excess permissions, or rebuilt from a stale or poisoned source. Queries are answered from an index whose contents were not reviewed, and the change is not visible to consuming applications. Integrity and availability of search results are compromised. | 1 | 1 | 1 |
| CCC.Vector.TH05 | Malformed Vectors Cause Failures or Incorrect Results | Vectors submitted with unexpected dimensionality, numeric type, or degenerate values such as NaN, infinity, zero magnitude under cosine similarity, or extreme magnitude may be accepted when validation is incomplete. Indexing or query components fail, or similarity scores are distorted for every query that encounters the vector. Availability of the service and integrity of results are impacted. | 1 | 1 | 1 |
| CCC.Vector.TH06 | Search Results are Manipulated via Approximate Search Error | Approximate nearest neighbor algorithms trade recall for speed and return results that vary with index structure and tuning parameters. Vectors may be positioned so that they are omitted from, or injected into, the approximate result set while an exact search would rank them differently, allowing records to evade similarity-based detection or to bias generated responses. Integrity of search results is compromised. | 2 | 1 | 1 |
| CCC.Vector.TH07 | Content is Disclosed to or Altered by an Embedding Model Endpoint | When the service generates embeddings by calling a model endpoint, raw content is transmitted to that endpoint and the vectors it returns are trusted without verification. A misconfigured, third-party, or substituted endpoint receives sensitive content, or returns embeddings that misrepresent the content. Confidentiality of the content and integrity of the index are compromised. | 0 | 1 | 1 |