Models and the systems around them can repeat confidential data from prompts, retrieval, training data or logs to the wrong person, and side channels can reveal more than the answer itself.
- Keep secrets and data the user may not see out of prompts, context and training sets, and minimize and redact what goes in.
- Enforce each user’s access rights on retrieval and tools before content reaches the model.
- Filter output for sensitive data, and limit what logs and traces keep.
- Measure memorization and membership-inference risk for models trained on sensitive data.
Guides on this site
Assess extraction, membership and inference risks
Define what an adversary can observe and measure privacy leakage without confusing a model’s output with proof about its training data.
Keep retrieval inside the user’s data boundary
Carry document permissions through ingestion, retrieval, caching and citations so a model never receives another tenant’s records.
Call a hosted LLM API safely
A code-first checklist for wiring up an OpenAI or Anthropic API call from a backend: keep keys server-side, redact PII before prompts and logs, choose provider retention settings deliberately, validate structured output and still treat it as untrusted, and set timeouts, retries and per-user cost caps.
Practice
References
- LLM02:2025 Sensitive Information Disclosure (OWASP, 2025 edition)
- OWASP Top 10 for LLM Applications 2026 (OWASP GenAI Security Project resource page)
List data from the publisher. A guide is listed as the main subject when the weakness or risk is what the guide is about, and as covering it when a dedicated section or worked example addresses it.