Glossary
Training Data Access
Training data access is the permission to use a dataset as input for training a model, evaluated separately from the question of who may later query the resulting trained model through an inference endpoint. A dataset that's sensitive enough to require careful access control during training doesn't stop being sensitive once a model has been trained on it — the model itself can retain traces of what it was trained on, which is exactly why access to the training process deserves its own scrutiny, not an assumption that once training finishes, the underlying data's sensitivity is somehow resolved.
Treating a training dataset's access controls as a one-time gate at the start of a project, rather than an ongoing concern, misses that the same sensitive data often gets reused for retraining, fine-tuning, or evaluation well after the original access decision was made — each of those later uses is a fresh instance of the same authorization question, not a foregone conclusion from the first approval.