Learning path · Enterprise Patterns & Governance · 86
Privacy and Data Handling
Policies for PII redaction, data residency, retention, and customer consent when sending text to models.
Why it matters
- Legal exposure if prompts leak regulated data to vendors.
- Determines self-host vs API and regional routing.
- Deletion requests must propagate through indexes.
Key ideas
- PII redaction
- Data residency
- Retention and deletion
Top resources
- 01DocsNIST
AI Risk Management Framework
Why this resource. Privacy and data as AI risk functions, not an afterthought policy PDF.
Covers in this concept
- data governance
- trustworthiness
- 02DocsOWASP
OWASP Top 10 for LLM Applications
Why this resource. Concrete leak and poisoning paths in LLM apps.
Covers in this concept
- sensitive info disclosure
- supply chain
Privacy engineering classifies fields, redacts before inference, and documents subprocessors. Use regional endpoints or local models for sensitive workloads. Encrypt vectors and metadata at rest; tie chunk ACLs to source systems. Run DPIAs when adding new tools that exfiltrate context to third parties. Review subprocessors whenever enabling new MCP servers or browser tools; data paths multiply faster than security inventories update. Automate redaction tests whenever new connectors ingest email or ticket bodies.
Updated 2026-08-09 · Full learning path