1:1 mentoring with Big Tech AI engineers
LLM & Agentic

Memory Compliance & Interview Prep

GDPR and privacy compliance for agent memory, plus FDE interview scenarios and deep-dive questions.

Last updated

Staff6 min readFirst readMemory Implementation & Patterns

After this section you can

  • Restate the four data-subject rights as engineering requirements a memory store must honor
  • Design deletion, export, and retention into the schema rather than bolting them on
  • Take a memory system through a pre-ship compliance checklist
04

Memory Compliance & Interview Prep

The moment an agent remembers users, it’s a data-processing system under GDPR — with deletion, correction, and consent obligations. The compliance you must build in, plus the FDE interview questions this whole track is really preparing you for.

THE UNCOMFORTABLE TRUTH

Long-term memory is a database of personal data that your AI wrote by itself, scattered across a vector store, distilled into facts nobody explicitly reviewed. Every privacy regulation that applies to a CRM applies to it — GDPR, CCPA, and friends — except it’s harder to audit because a model decided what to save. If you add per-user memory and can’t answer “show me everything you know about this person, and delete it on request,” you haven’t built a feature, you’ve built a liability. Compliance is not a later phase; it’s a design constraint on the store itself.

The four rights an agent-memory store must be able to honor
ERASURE Art. 17 “forget me” delete ALL memories for a user, every store needs: user-id index ACCESS Art. 15 “what do you know?” export everything about a user, human-readable needs: queryable by user RECTIFICATION Art. 16 “that’s wrong” correct or overwrite a stored fact on request needs: update-in-place CONSENT + TTL Art. 5, 6 “why & how long?” store only with a basis; expire on retention TTL needs: TTL + provenance

Related

More in LLM & Agentic

Get full access to all 74+ sections with code examples, diagrams, and interactive animations.

Unlock Premium