French PII Detection and Redaction
Sanitext detects and redacts PII in French text, including names, emails, French phone numbers (+33), postal addresses, dates, and account numbers. The model is trained multilingually, so French is a native language, not a bolt-on. English-only tools like AWS Comprehend miss French-specific formats and accents.
Most PII tools are built for English first and treat French as an afterthought. They trip on accented names, the +33 phone format, and French address layouts. Sanitext runs an open-weights model trained across 30+ languages, so French text is handled the same way English is. You send French text, you get back labeled spans or masked output. No language flag to set, no separate French model to pick.
French PII coverage: Sanitext vs English-first tools
| French feature | Sanitext | English-only tools (e.g. AWS Comprehend) |
|---|---|---|
| Accented names (Hélène, François) | Detected natively | Often missed or mislabeled |
| +33 / 06 phone formats | PHONE label, both formats | Inconsistent without config |
| French address layouts | ADDRESS label | Tuned for US-style addresses |
| Language setup | Automatic, no flag | Per-language model or flag |
| Data sent to third-party LLM | No, self-hosted on Cloudflare | Often a managed cloud service |
| Pricing | Flat monthly + cheap overage | Metered per character |
What PII does Sanitext detect in French?
Sanitext reads French text and tags personal data the same way it does in English. It handles accented first and last names (think Hélène, François,Ézéchiel), French email addresses, mobile and landline numbers in +33 or 0X format, postal addresses with French street types, dates in day-month-year order, URLs, account and card numbers, and secrets like API keys and tokens.
Every match comes back with a label, a confidence score, and byte offsets into the original string. Byte offsets matter for French because accented characters use multiple bytes in UTF-8, and you need exact positions to mask the right text.
- —FIRSTNAME and LASTNAME, including accented forms (é, è, ç, ï, â)
- —EMAIL, like jean@exemple.fr
- —PHONE, French formats: +33 6 12 34 56 78 and 06 12 34 56 78
- —ADDRESS, French postal layouts with rue, avenue, boulevard
- —DATE, day-month-year written forms
- —URL, ACCOUNT, and SECRET (API keys, tokens)
- —PREFIX, like Dr. or M. before a name
Worked example: French sentence to redacted output
Take this input string:
"Bonjour, je m'appelle Jean Dupont, mon email est jean@exemple.fr et mon numéro est +33 6 12 34 56 78."
Call POST /v1/redact and Sanitext returns the same sentence with each piece of PII masked by its label. The grammar and structure stay intact, so the text is still readable and safe to log, store, or send to a third-party LLM.
Redacted output:
"Bonjour, je m'appelle [FIRSTNAME] [LASTNAME], mon email est [EMAIL] et mon numéro est [PHONE]."
If you call POST /v1/detect instead, you get the raw spans (label, score, byte offsets, and the matched text) so you can apply your own masking, hashing, or tokenization logic.
How accurate is French detection?
French is a first-class language in the model, not a translation layer. Detection quality on common entities like names, emails, phones, and URLs is strong because these patterns are stable across languages. Structured data with clear shapes, like emails and +33 numbers, is the most reliable.
We are honest about the hard cases. Free-text French addresses without a postal code, rare or foreign-origin surnames, and names that double as common French words can be missed or mislabeled. Detection runs on probabilities, so review the confidence score and set a threshold that fits your risk level.
Sanitext is a redaction and data-minimization aid. It is not an anonymization guarantee and not proof of GDPR compliance. Treat it as a strong first pass, then add human review or a second control for high-stakes data.
Where does French data get processed?
Sanitext runs on our own Cloudflare infrastructure using an open-weights model we host ourselves. Your French text never goes to a third-party LLM provider like OpenAI or AWS. We log counts and timings only, not raw text.
For EU customers handling French personal data, this matters under GDPR. Enterprise plans include EU data residency and a signed DPA, so you can keep processing inside the EU and document the data flow for your records. Pricing is flat per month with cheap overage, not metered per token, because our AI cost is effectively zero.
bash
curl -X POST https://api.sanitext.app/v1/redact \
-H "Authorization: Bearer $SANITEXT_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"text": "Bonjour, je m'\''appelle Jean Dupont, mon email est jean@exemple.fr et mon numéro est +33 6 12 34 56 78."
}'
# Response
# {
# "redacted": "Bonjour, je m'\''appelle [FIRSTNAME] [LASTNAME], mon email est [EMAIL] et mon numéro est [PHONE]."
# }FAQ
Do I need to set a language flag for French?+
No. Sanitext auto-handles French because the model is trained on 30+ languages at once. You send French text to the same /v1/detect or /v1/redact endpoint you use for English. There is no French-specific model or parameter to configure.
Does Sanitext handle accented French names?+
Yes. Names with accents like é, è, ç, ï, and â (for example Hélène or François) are detected as FIRSTNAME or LASTNAME. Spans return byte offsets, which correctly account for multi-byte UTF-8 accented characters so you mask the exact text.
Can Sanitext redact French phone numbers like +33 6 12 34 56 78?+
Yes. It detects French mobile and landline numbers in international +33 format and national 0X format, with or without spaces. These are returned with the PHONE label and masked as [PHONE] when you call /v1/redact.
Is Sanitext enough for GDPR compliance on French data?+
No tool alone makes you GDPR compliant. Sanitext is a redaction and data-minimization aid that reduces PII exposure. It is not an anonymization guarantee. Pair it with human review and your own controls, and use Enterprise EU residency plus a DPA for documentation.
Redact French PII in 60 seconds
Sign up, get an API key, and run your first French redaction with a single curl. Free 300K characters, no card required. Try French text live in the playground or read the docs.
Get your free API key