Redact whole document sets for research and AI

Minimum order: 10,000 patient records. PHI Mask Batch offers redaction or supports de-identification workflows across patient records, legal files and financial documents. Run it on-prem, or our team does it for you or with you.

HIPAA-compatible offline, or under your BAA and NDA

On-prem

Run offline, PHI Mask Batch is HIPAA-compatible. Your files are masked in your environment and never sent to us, so there is no BAA to sign, and without a signed BAA we cannot claim HIPAA compliance.

It runs inside the controls you already audit, so it is also GDPR-compatible, PIPEDA-compatible, PHIPA-compatible, SOC 2-compatible and ISO 27001-compatible.

Your own staff review the sample, so nobody from PHI Mask sees a file.

With our team

Our team sets the rules, runs the batch and checks the sample, for you or alongside your staff. In your environment or ours, under your BAA and NDA.

In Canada, we sign your agent or information manager agreement. In the EU, your data processing agreement.

You keep the key and the sign-off.

Both routes run the same software. You pick the route for each batch.

PHI Mask holds no SOC 2 report and no ISO/IEC 27001 certificate. Privacy and security review

Who it's for

  • A clinic partner holding back charts

    They run it on their side. You receive redacted copies.

  • A study waiting on ethics or IRB review

    Under TCPS 2, secondary use of non-identifiable information needs ethics review but not consent.

  • A model that learns from documents

    Names and numbers are redacted before training. The language stays.

  • A law firm putting case files into AI

    Client names, addresses and account numbers are redacted before review.

  • A finance team sharing statements

    Names and account numbers are redacted before an auditor, vendor or model sees them.

  • An AI vendor pilot

    Give the vendor redacted copies. The originals stay where they are.

Each document stays readable

We start from a rule set for health, legal or financial documents. Your privacy office, expert or counsel can change it.

Fictional note, after masking. Dates shifted.

[patient-0412], born 1980. Health number [health].

Seen 2026-03-02. Harvey-Bradshaw Index 5. Weight stable at 68 kg.

MR enterography on 2026-06-06 showed 12 cm of active terminal ileal inflammation.

Works at [employer]. Lives with partner [relative-1] in [place].

Names
One random code per patient, across every file
Birth dates
The year stays. Age 90 and over becomes one band
Visit dates
Shifted per patient, so the intervals stay
Postal codes
First three characters, or none for small areas
Family, work, town
Relatives, employers and home towns are redacted
Subject matter
Drugs, doses, case law and amounts stay

What you get back

Redacted copies
The same folder structure, with names taken out of file and folder names. Redacted text for your models, page-image PDFs for people.
Your key
One key maps each random code back to a person. It stays with you, apart from the people using the copies, and you can delete it.
A release report
Counts, hashes, the rules applied and every miss the sample review found, for whoever signs off.
Release report Example
Files in10,000
Masked9,964
Held for a person36
Pages read212,480
Scanned pages31%
Names418,202
Dates96,114
Record numbers61,870
Phones and emails20,913
Addresses12,455
Sample reviewed twice3,000 pages
Misses found in that sampleListed by type
Rule setv3 · sha256 41ab…07c2
Buildsha256 9f2c…e41a

Counts and hashes only. No patient text.

How a batch runs

  1. Count

    A counts-only inventory of formats, pages and scans.

  2. Dry run

    A fictional set runs first, and your team reads a real report.

  3. Set the rules

    Your privacy office, expert or counsel sets what is redacted.

  4. Redact

    Every file runs on-prem, or with our team.

  5. Review

    Your reviewers, or ours, check flagged files and a 3,000-page sample. Plan on about 200 reviewer hours. A further batch on the same rules samples 600 pages.

  6. Sign off

    Your privacy office, expert, ethics board or counsel reads the release report.

Common questions

Who reads the files?

On-prem, your own staff do, so nobody from PHI Mask sees a file. When our team does it, our reviewers work under your BAA and NDA.

Is the output de-identified?

That decision belongs to your privacy office, expert or ethics board. The release report gives them the counts, hashes and sample findings to decide. In Canada, it turns on who holds the key.

Who writes the expert letter?

An independent expert you choose. The batch rules, the release report and the sample findings are their evidence.

What do you sign?

When our team does it: your BAA and NDA, your agent or information manager agreement in Canada, your data processing agreement in the EU. On-prem, no BAA, because your files never reach us.

What does it run on?

On-prem: one offline container on your server, your cloud or your partner's machine, with no outbound connections. With our team: your environment or ours.

Does it use an LLM or generative AI?

No. A standard batch uses pattern matching and OCR, with no LLM or generative AI.

Does it read scans?

Yes. Scans are read page by page. Handwriting is read on a best-effort basis, and those pages are held for a person.

How long does a batch take?

The sample review sets the date, and we agree it when we scope.

Priced by the page

Minimum order: 10,000 patient records. Scans included. 10,000 five-page files are 50,000 pages: US$4,000 on-prem.

On-premWith our team
Per 1,000 pagesUS$80US$160
Set-up, once per rule setUS$2,500US$2,500
First sample reviewYour staffUS$10,000

Over a million pages, we quote volume pricing. Under 10,000 patient records? For individual files: PHI Mask Professional masks them in your browser

Start with counts, not files

Tell us the record count, file types and who runs the batch. We run a fictional set first, before any real file goes in.

Scope a batch