Source: Presidio documentation

Open-source PII detection and de-identification SDK: analyzer (NER + regex + checksums + context), anonymizer operators, image and structured-data redaction, with an explicit no-guarantee warning.

Contents
  • What: project documentation (home and anonymizer pages)
  • Publisher: Presidio project, originally Microsoft. The site says it is now community-owned under “Data Privacy Stack”.
  • Date: accessed 2026-11-14
  • Original: presidio.dataprivacystack.org · Anonymizer

Summary

Presidio is the usual open-source building block for finding and masking PII in text, images and tables before data reaches an LLM, a log, or a training set.

Key claims

  • Components: Analyzer (detects PII in text), Anonymizer (applies operators), Image Redactor (OCR then redaction), Structured (tabular and semi-structured data).
  • Recognizers combine NER models, regular expressions, rule logic and checksums, using surrounding context, across several languages. Custom recognizers can be added.
  • Anonymizer operators: replace (defaults to an <ENTITY_TYPE> placeholder), redact, mask (chars_to_mask, masking_char, from_end), hash (sha256 or sha512, salted; minimum salt 16 bytes; a fixed salt is needed for consistent hashes), encrypt (keyed), custom (lambda returning a string), surrogate_ahds (surrogates via Azure Health Data Services, needs the [ahds] extra), keep.
  • Only encrypt is reversible: decrypt with the same key.
  • The docs warn that automated detection cannot guarantee it finds all sensitive data, and recommend extra protections.

Feeds

Enterprise LLM deployment

Suggest changes →