During an AI model audit, the auditor requests access to the model's training data but is told it was deleted after training to reduce storage costs. What risk does this create for ongoing model governance?

B is correct because deleting training data creates multiple governance failures: training cannot be reproduced or audited, bias investigations cannot trace root causes to specific data subsets, and data subject rights (e.g., right to erasure verification) cannot be fully honored. A is incorrect — training data is not cached for inference. C is a potential but not certain consequence (GDPR minimization might actually require deletion). D is a practical retraining limitation, not the primary governance risk.