AI for healthcare teams, without sending patient identities
Discharge summaries, referral letters, lab reports, claims files: AI can save clinicians and administrators hours. Hidden Tokens lets them use it while patient identities stay on the computer.
- Patient names, record numbers, insured and practitioner identifiers, social security numbers and batch numbers are coded on your computer before the AI reads the document.
- Specialties, procedures, tests, common medicines and health bodies stay readable, so the AI can still work on the medical content.
- The letter or summary comes back with the real names restored, on your computer.
- Health data is a special category under the GDPR. Hidden Tokens reduces what reaches the AI; it does not by itself make a use lawful, nor does it certify HIPAA or any other compliance.
Why patient data needs more than a policy
Health data is among the most protected categories of personal data in every major privacy law. Yet the documents where AI helps most (long reports, correspondence, claims) are also the ones dense with identifiers. Asking staff not to paste them into a chatbot rarely works; making sure that what they paste identifies no one does.
What the healthcare pack codes, and what it keeps readable
| Coded | Kept readable |
|---|---|
| Patient, relative and practitioner names | Specialties, departments and procedures |
| Patient record numbers | Tests, results and medical vocabulary |
| Insured-person and social security numbers | Common medicines (international non-proprietary names) |
| Practitioner identifiers | Health bodies and schemes |
| Batch numbers, addresses, phone numbers | Dates of care and durations |
Typical uses
- Clinics and hospitals: drafting referral letters and discharge summaries from notes.
- Laboratories: summarising result series, tracing a batch number across reports.
- Health insurers: triaging claims files and drafting responses.
Proof for your DPO
A local page lists every response the AI received and re-checks it against your key; the expected result is zero real values. Your data protection officer can verify signed proof pages in a browser without uploading anything. See pseudonymisation vs anonymisation for why the key's location matters.
Frequently asked questions
Can the AI still understand a medical report if the patient is coded?
Yes. Only identifiers are coded. Diagnoses, procedures, test results, medicines and dates stay readable, and the patient keeps the same code across all their documents.
Which healthcare identifiers does the pack recognise?
Patient record numbers, insured-person numbers, practitioner identifiers and medicine or device batch numbers, in the formats used in France and Morocco (for example RPPS, ADELI, INPE, AMO, CNOPS, CNSS). The French social security number is recognised by the core engine. Other formats can be added as custom rules.
Does it handle scanned documents?
Yes. Hidden Tokens reads scans and can redact images, so a scanned referral letter is coded like a typed one.