Notesfoundations🌱 seedling
Right to Erasure
Also: Right to be Forgotten · Data Deletion Request · GDPR Article 17
The right to erasure lets people have their personal data deleted without undue delay when it's no longer needed or they withdraw consent. For engineers it is a distributed-systems problem: the data has been copied into many stores, and every copy must go.
Why it exists
People leave services, withdraw consent, or object to processing. Under GDPR Article 17 the controller must erase their data, answer within one month (extendable by two for complex cases) and tell everyone the data was disclosed to unless that's impossible or disproportionate (Article 19). Exceptions include legal obligations, legal claims, public health, research and freedom of expression.
How it works
- Inventory: every store holding personal data (caches, search indexes, logs, warehouses, event streams, backups, ML features, third parties) registers what it holds, keyed by user ID.
- Fan-out job: an erasure request becomes a durable workflow that calls every registered store, collects confirmations, retries laggards and alarms when one stays stuck.
- Derived Data: indexes and caches rebuilt from a source must be purged or rebuilt after the source deletes.
- Backups: expire on schedule; restores re-apply pending deletions.
- Undeletable stores (append-only logs, archives): Crypto-Shredding, or tombstones in compacted topics.
- Models: regulators have ordered deletion of models trained on improperly obtained data.
Trade-offs
| Gains | Costs |
|---|---|
| legal compliance, user trust | registration discipline for every new store |
| a complete data inventory as a side effect | engineering a reliable fan-out workflow |
| fewer stale copies lying around | reversibility windows complicate "deleted" |
In practice
- Meta DELF: developers annotate data types and graph edges with deletion behaviour; deletions run asynchronously with retries and a restoration log. It handles billions of deletions a day and its checks found thousands of missing annotations.
- FTC: Everalbum (2021) and WW/Kurbo (2022) were ordered to delete models built from improperly collected data.
Pitfalls
- Deleting the primary row and replying "done".
- A new store added to the architecture but never registered with deletion.
Connections
- Part of
- Data Ethics
- Contains
- Crypto-Shredding
Recall
These cards are also in your review deck, scheduled with spaced repetition.