Data retention sets how long information is stored to meet operational, legal, or policy requirements. For AI systems, deletion may need to reach beyond the main database: agents can leave data in working memory, logs, saved outputs, persistent memory, and downstream services. section 3.3.3 section 5.6.3
Turn retention requirements into operational controls:
- Define timelines for each store and apply short TTLs and automated purges to working memory, logs, and caches. section 5.6.3
- Make stores searchable and purgeable; route erasure requests across the main application, agent memory, tool logs, outputs, and services that received the data. section 5.6.3
- For RAG, manage retention for indexed documents and embeddings, and remove both when required. section 5.2.4
- Track deletion evidence rather than relying on contract language alone; determine separately whether data already used for training requires a model-level response. section 3.3.3 section 5.6.3
Aligning a vendor’s retention period
A healthcare image feature faces a vendor default of 30 days against a seven-day organizational policy. Renegotiate the terms, then verify scheduled deletion through storage logs or vendor deletion records. section 3.3.3
