De-identify clinical notes before you paste them into an AI
This preset turns on the detectors that map to HIPAA’s Safe Harbor list — including all dates, which are off everywhere else on the site because ordinary writing is full of harmless ones. Medical record numbers, account numbers, phone numbers, emails, addresses, postcodes and names are all on.
Read the caveat before you rely on it: this tool removes identifiers it can recognise. It cannot certify that a note is de-identified, and de-identification under HIPAA is a legal standard, not a feature.
Your text
Safe to paste
Preset: the Safe Harbor-shaped detector set, including every date.
Mapping — 0 values (only visible in this tab)
| Placeholder | Real value | Uses |
|---|
Check a few rows before you paste — this is the fastest way to spot a detector that grabbed the wrong thing. Nothing here is stored or sent anywhere.
What to look for — 30 detectors, custom terms, exceptions
People & organisations
Government & health IDs
Financial
Technical & secrets
Location
Step 3 — paste the AI’s answer here to put the real values back
The model replies using the placeholders. Paste its answer below and every
[NAME_1] becomes the real name again — using the mapping held in this tab, which
is the only copy that exists.
How it works
- Open this page — the detectors are already set for “Medical notes”.
- Paste your text, or drop a .txt, .md, .csv, .json or .log file onto the panel.
- Check the highlights: every match is coloured by type, and one click keeps a type as-is.
- Copy the safe version and paste it into the AI tool of your choice.
- Paste the model’s reply back into step 3 to restore the real values.
What Safe Harbor actually requires
The Safe Harbor method in 45 CFR 164.514(b)(2) lists eighteen identifier types that must be removed: names; geographic subdivisions smaller than a state (including street address, city, county and, with a narrow exception, ZIP code); all elements of dates except the year, plus any age over 89; telephone and fax numbers; email addresses; Social Security numbers; medical record numbers; health plan beneficiary numbers; account numbers; certificate and licence numbers; vehicle and device identifiers; URLs and IP addresses; biometric identifiers; full-face photographs; and any other unique identifying number, characteristic or code.
There is a second condition people forget: the covered entity must also have no actual knowledge that the remaining information could identify the individual, alone or in combination with other data. A note that says “the only patient in our practice with this rare condition” fails that test even if every name is gone. HIPAA’s other route, expert determination, exists precisely because pattern removal alone often is not enough.
How the preset maps onto that list — and where it does not
Well covered: names, addresses and postcodes, dates, phone numbers, email addresses, SSNs, medical record numbers, account and licence numbers, IP addresses. Ages over 89 are not detected as such, and the “any other unique characteristic” catch-all is by definition unpatternable — that is what the custom-terms box is for.
Clinical narrative is also full of near-identifiers a regular expression will never see: an unusual occupation, a named employer, a rare procedure on a specific date, the name of a small clinic. If the note is going to leave your environment, read it once with the question “could someone who knows this person recognise them here?” — that question, not the detector count, is the standard.
Tips
- For research datasets, remember Safe Harbor keeps only the year of a date and requires ages over 89 to be aggregated.
- UK clinicians: switch on the NHS number detector — it is off by default because a valid NHS number is ten digits and looks exactly like a US phone number.
- Check whether your organisation has a sanctioned AI route with a business associate agreement in place before using a public chatbot at all.
Detectors are pattern-based and imperfect: read the highlighted output before you paste it anywhere, and confirm your own confidentiality obligations against the current rules for your profession, employer or jurisdiction. Compiled 2026-07-29.
Frequently asked questions
Does using this make me HIPAA-compliant?
No. It is a mechanical aid for removing identifiers. Compliance depends on your role under HIPAA, your agreements, the minimum-necessary rule and the actual-knowledge condition in the regulation. Treat the output as a draft that a human must review.
Why are all dates on here but off everywhere else?
Because Safe Harbor treats dates as identifiers: admission, discharge, surgery and birth dates all have to go, keeping only the year. In ordinary business text, redacting every date makes the content useless, so it is opt-in there and default here.
Is it safe to put patient data into a chatbot at all after scrubbing?
That depends on the deployment, not on this tool. Consumer chatbot plans train on conversations unless you opt out and do not offer a business associate agreement; enterprise and API tiers typically do neither train by default nor refuse a BAA. De-identified text is outside HIPAA’s scope, but the “de-identified” judgement is yours to make and to document.