Pseudonymization replaces identifying information with pseudonyms while retaining the possibility of reidentification through separately held additional information. In AI applications, it protects chat histories used for analysis or model training without treating privacy as an afterthought. section 9.5 section 9.7
Apply pseudonymization as part of a broader data-protection design:
- Keep identifying information out of prompt queries whenever possible. section 9.5
- Avoid logging or sending sensitive details to LLM providers whose PII policies do not meet the application’s requirements. section 9.5
- Combine pseudonymization with access control, retention policies, encryption, logging, and auditing. section 9.5 section 9.6.2
- Evaluate pseudonymization alongside provider privacy policies and regulatory obligations such as GDPR and HIPAA. section 9.6.2
Chat-history analysis
Before using stored user conversations for model training or analysis, replace identifying details with pseudonyms and keep the additional reidentification information separately held. Review whether prompts, logs, and external LLM requests still contain PII. section 9.5
