Access, correction, and deletion requests become difficult when personal data has spread across prompts, logs, retrieval indexes, model parameters, outputs, vendors, and backups. section 5.0
The governance challenge is to honor individual rights despite that spread. Machine unlearning remains an active research area, not a reliable enterprise capability, so organizations should explain what they can deliver and avoid promising complete model erasure. section 5.1
Build operational controls around those limitations:
- Create deletion-aware pipelines before data enters systems where removal is difficult. section 5.1
- Keep records, tests, contracts, deletion logs, and system controls to show where personal data went and what the organization did. section 5.0
- Test and document deletion across weights and embeddings, and state clearly where current capabilities fall short. section 5.1
