AI systems ingest large volumes of documents and logs, and much of that content contains personal data. The pages in this section cover the techniques used to find that data, transform it so it is safer to process, release aggregates without exposing individuals, and delete it on request, including from vector stores used by RAG.
Protecting personal data in AI pipelines: PII detection and redaction, pseudonymization, k-anonymity, differential privacy and right-to-erasure.
A firm running document intelligence across hundreds of matters will eventually want cross-matter analytics: "how many contracts reference arbitration…
Releasing a dataset — a discovery set, a training corpus, a research extract — is different from running a query against one. Once the rows leave your trust…
In a legal document-intelligence platform, the single riskiest moment in the data lifecycle is ingest . Once a document has been split, embedded, and written…
GDPR Article 17 — the right to erasure, colloquially "right to be forgotten" — requires that, when a data subject withdraws consent or objects to proc
The regex + spaCy pipeline on the PII redaction page is fast and transparent, but extending it to new entity types (bar numbers, docket numbers, FDA…
Three terms are used almost interchangeably in marketing copy and mean meaningfully different things under GDPR, HIPAA, and CCPA. Getting the distinction…
Detection comes first: rule-based recognizers and NER models such as Presidio locate names, identifiers and contact details in structured and unstructured data. Transformation then removes or replaces them: redaction deletes, tokenization swaps values for reversible tokens held in a vault, pseudonymization replaces identifiers consistently, and anonymization aims to make re-identification impossible. When releasing statistics, k-anonymity and differential privacy limit what can be learned about any one person. Finally, erasure requests must reach every copy, including embeddings and chunk stores.
| Technique | Reversible | Still personal data under GDPR | Use when |
|---|---|---|---|
| Redaction | No | Depends on what remains | Sharing documents, prompts to external LLMs |
| Tokenization | Yes, via vault | Yes | Systems need the real value later |
| Pseudonymization | With the key | Yes | Joining records without exposing identity |
| k-anonymity / l-diversity | No | Possibly | Releasing record-level tables |
| Differential privacy | No | Aggregates generally not | Publishing counts and statistics |