Private data detection

The translation API checks input terms for private or personal data before returning model translations. The check is designed to prevent patient or professional identifiers from being sent through the translation workflow.

Detection behavior

The API uses Presidio-style entity detection on each submitted source term. When sensitive content is found, the request returns HTTP 422 with the standard message:

{
  "error": "Unprocessed Entity: term includes sensitive data",
  "message": "term includes sensitive data"
}

For multi-term requests, each term is checked independently. Terms without private data can still be translated by the batch translation path, while terms with private data are blocked and receive the private-data message in their output. If at least one term is blocked, the HTTP status is 422.

Remote API summary

The following synthetic examples were tested against the remote mTLS endpoint https://anstranslation2.ddns.net/translate. All names and identifiers are test values, not real patient data.

Category	Requested terms	HTTP status	Result
Safe medical control	heart failure; respiratory distress	200	All translated
Name, first name, maiden name	Patient John Doe has heart failure; Jane Smith nee Miller reports respiratory distress	422	All blocked
Email address	john.doe@example.com reports chronic kidney disease	422	Blocked
Phone number	Patient callback 555-123-4567 for diabetes mellitus follow-up	422	Blocked
French social security number / NIR	Patient NIR 180027512345678 treated for asthma	200	Translated
US SSN-style social security number	Patient SSN 123-45-6789 has hypertension	200	Translated
Patient identifier / IPP	IPP 987654321 patient diagnosed with acute bronchitis	200	Translated
Mutuelle / complementary insurance number	Mutuelle number MGEN-99887766 for patient with migraine	422	Blocked
RPPS / ADELI professional identifier	Dr Martin RPPS 10101234567 follows patient for heart failure; ADELI 751234567 associated with patient diabetes care	422	One blocked, one translated
Messaging identifier	Skype john.doe.med for respiratory distress consultation; WhatsApp +33 6 12 34 56 78 for oncology follow-up	200	All translated
Date and place of birth	John Doe born 1980-02-12 in Paris has chronic asthma	422	Blocked
Passport / identity card number	Passport 123456789 for patient with pneumonia; ID card AB123456 patient with hypertension	200	All translated
Driving licence / fiscal number	Driver license D1234567 patient with diabetes; Tax SPI 1234567890123 patient with renal failure	200	All translated
IBAN, RIB, credit card	IBAN FR7630006000011234567890189 for reimbursement of cardiac care; Credit card 4111111111111111 for medical invoice	422	All blocked
Bank account number	Bank account 123456789012 for reimbursement of diabetes care	200	Translated
Mixed safe and sensitive batch	heart failure; john.doe@example.com reports chronic asthma	422	One translated, one blocked
Disease names resembling people names	Crohn's disease; Krohn disease; Parkinson disease; Alzheimer disease; Hodgkin lymphoma; Graves disease	422	Four translated, two blocked
Eponym medical syndromes	Down syndrome; Turner syndrome; Marfan syndrome; Guillain Barre syndrome; Asperger syndrome	422	Three translated, two blocked
Medical terms with initials and codes	type 2 diabetes mellitus; HLA-B27 associated uveitis; BRCA1 mutation carrier; COVID-19 pneumonia; ICD-10 code I50 heart failure	200	All translated
Mixed disease control and explicit person name	Crohn's disease; Patient Alice Martin has Graves disease	422	One translated, one blocked

Detailed term results

This section lists only the submitted terms and the observed result for each term. It intentionally omits the full request and response bodies.

Requested term	Result	Test group
heart failure	Translated	Safe medical control
respiratory distress	Translated	Safe medical control
Patient John Doe has heart failure	Blocked	Name, first name, maiden name
Jane Smith nee Miller reports respiratory distress	Blocked	Name, first name, maiden name
john.doe@example.com reports chronic kidney disease	Blocked	Email address
Patient callback 555-123-4567 for diabetes mellitus follow-up	Blocked	Phone number
Patient NIR 180027512345678 treated for asthma	Translated	French social security number / NIR
Patient SSN 123-45-6789 has hypertension	Translated	US SSN-style social security number
IPP 987654321 patient diagnosed with acute bronchitis	Translated	Patient identifier / IPP
Mutuelle number MGEN-99887766 for patient with migraine	Blocked	Mutuelle / complementary insurance number
Dr Martin RPPS 10101234567 follows patient for heart failure	Blocked	RPPS / ADELI professional identifier
ADELI 751234567 associated with patient diabetes care	Translated	RPPS / ADELI professional identifier
Skype john.doe.med for respiratory distress consultation	Translated	Messaging identifier
WhatsApp +33 6 12 34 56 78 for oncology follow-up	Translated	Messaging identifier
John Doe born 1980-02-12 in Paris has chronic asthma	Blocked	Date and place of birth
Passport 123456789 for patient with pneumonia	Translated	Passport / identity card number
ID card AB123456 patient with hypertension	Translated	Passport / identity card number
Driver license D1234567 patient with diabetes	Translated	Driving licence / fiscal number
Tax SPI 1234567890123 patient with renal failure	Translated	Driving licence / fiscal number
IBAN FR7630006000011234567890189 for reimbursement of cardiac care	Blocked	IBAN, RIB, credit card
Credit card 4111111111111111 for medical invoice	Blocked	IBAN, RIB, credit card
Bank account 123456789012 for reimbursement of diabetes care	Translated	Bank account number
heart failure	Translated	Mixed safe and sensitive batch
john.doe@example.com reports chronic asthma	Blocked	Mixed safe and sensitive batch
Crohn's disease	Translated	Disease names resembling people names
Krohn disease	Blocked	Disease names resembling people names
Parkinson disease	Translated	Disease names resembling people names
Alzheimer disease	Translated	Disease names resembling people names
Hodgkin lymphoma	Translated	Disease names resembling people names
Graves disease	Blocked	Disease names resembling people names
Down syndrome	Translated	Eponym medical syndromes
Turner syndrome	Translated	Eponym medical syndromes
Marfan syndrome	Blocked	Eponym medical syndromes
Guillain Barre syndrome	Blocked	Eponym medical syndromes
Asperger syndrome	Translated	Eponym medical syndromes
type 2 diabetes mellitus	Translated	Medical terms with initials and codes
HLA-B27 associated uveitis	Translated	Medical terms with initials and codes
BRCA1 mutation carrier	Translated	Medical terms with initials and codes
COVID-19 pneumonia	Translated	Medical terms with initials and codes
ICD-10 code I50 heart failure	Translated	Medical terms with initials and codes
Crohn's disease	Translated	Mixed disease control and explicit person name
Patient Alice Martin has Graves disease	Blocked	Mixed disease control and explicit person name

Interpretation

The detector blocks obvious private data such as synthetic names, email addresses, phone numbers, birth information, and banking examples. Some healthcare-specific identifier formats were not blocked in this remote test, including the tested NIR-like value, IPP-like value, passport/ID values, driver/fiscal values, and standalone bank account number.

The disease-name tests show an important false-positive risk: medical eponyms or misspelled disease names can be detected as PERSON. For production use, these cases should be handled with a terminology allowlist before private-data blocking, or reviewed manually when a clinical term is blocked as PERSON.