The Data Engine.

Most AI data access runs on a shared service account with read on everything. The model inherits whatever that account can see, retrieval pulls back rows no one vetted, and sensitive fields land in a prompt that gets forwarded to a third party. The leak rarely happens in the model - it happens upstream, in the access nobody scoped.

The Data Engine sits between your data and your models and makes access answer to policy: gated per request, masked before it leaves the boundary, and recorded as evidence. AI sees exactly what the asker is entitled to - and nothing else.

The Data Engine resolves every access decision against the same identity, purpose, and jurisdiction that Access Identity binds to each request, and enforces it inline through the same control layer that governs prompts and outputs. One policy model, one audit trail, and one traceable lineage from source record to prompt.

When real data should never be in play at all - in development, testing, or sharing across teams - the Synthetic Engine stands in with statistically faithful data that contains none of your records.

It is also where data governance stops being a catalog and becomes enforcement. Whether a model reaches a warehouse, a lake, or a live table, the same identity- and purpose-bound rules apply on the read - which is what turns data loss prevention from a network afterthought into a control on the AI data path itself.