Glossary
The personal data Anoni detects.
Every type below is found and replaced locally, by the French models or by rules.
- Person
- A first name, last name, or full name of a real person. The French model reads the sentence and replaces the name with a believable stand-in.
- Organization
- A company, firm, public body, or association named in the text. Anoni swaps it for a plausible organization name.
- Location
- A city, region, or place name that points to where someone is. Anoni replaces it with another real-sounding location.
- Address
- A French postal address: street number, street name, postcode, city. Anoni rewrites the whole address as a coherent fake one.
- An email address, in any common format. Anoni replaces it with a syntactically valid fake address.
- Phone
- A phone number, French or international. Anoni swaps it for a number of the same shape.
- NIR
- The French social-security number, the 15-digit identifier on a carte Vitale. A rule spots the pattern and replaces it with a structurally valid fake.
- SIRET / SIREN
- The French company identifier: nine digits for SIREN, fourteen for SIRET. A rule catches the number and substitutes one of the right length.
- IBAN
- A bank account number, including the French FR IBAN with its checksum. A rule validates the checksum, then replaces it with a fake IBAN that still checks out.
- URL
- A web link that can reveal a profile, a file, or an internal tool. Anoni replaces the link with a neutral fake URL.
- Date
- A date, in digits or written out in words. Anoni shifts it to another plausible date in the same format.
- Secret
- A technical key or token, the kind that should never sit in a shared document. A rule flags it and replaces it with a fake of the same shape.
- Custom
- Your own dictionary: client names, internal references, project codenames. You add the terms, and Anoni replaces every match the same way.
Two methods, one pass
Formatted identifiers come from rules. NIR, SIRET, IBAN and a phone number all follow a known shape. Semantic entities come from the model. A name or an organization only makes sense in context. Running both is what covers the most.
The honest limit
No detector catches everything. On a sensitive document, read the result before you share it. The tool finds the data. You keep the last word.
See the measured detection benchmark, the anonymize-before-AI guide, or download Anoni.