What PII redaction is, and how it differs from PII masking
Personally identifiable information (PII) is anything that identifies a person on its own or in combination: names, addresses, phone numbers, government and account numbers, dates of birth, signatures, faces and voices. Health information about a person (PHI) and business identifiers such as policy or claim numbers are usually handled the same way.
- PII redaction removes those values from the output, so the recipient cannot see or recover them.
- PII masking hides or replaces them, for example with asterisks, tokens or realistic synthetic values. Consistent replacement keeps a document readable and useful for testing: the same customer gets the same substitute name in every file.
- PII flagging is not redaction. A tool that marks sensitive values in extracted data has not removed them from the source document.
Both redaction and masking should leave the source values unrecoverable from the shared file. For how these terms relate to anonymization and pseudonymization, see data anonymization tools.
True removal vs black boxes
Most redaction failures are not missed detections. They are values that were covered instead of removed.
- A drawn rectangle or highlight is an annotation. The text underneath stays in the file and can be copied, searched or extracted.
- A PDF editor's redaction mark removes content only once the redactions are applied and the file is saved. A file sent before that step still holds everything.
- Earlier revisions can survive in a PDF saved incrementally, along with comments, form fields, bookmarks and embedded files.
- Partial characters at the edge of a box on a scan can be enough to read a number back.
True removal replaces the content itself: characters are deleted from the text layer, pixels are overwritten in the page image, and metadata is cleaned. Replacement with a consistent synthetic value is also removal, as long as the original is gone.
On-premises vs cloud PII redaction
Where redaction runs decides who else touches the unredacted file. With a cloud service, the original document travels to the provider before any PII is removed. With an on-premises or air-gapped deployment, it never leaves your network.
| Question | Cloud service | On-premises or air-gapped |
|---|
| Where does the unredacted file go? | To the provider's cloud | Nowhere outside your environment |
|---|
| What needs review? | The provider's retention, training and sub-processor terms | Your own infrastructure, access control and patching |
|---|
| What does it suit? | Data you are allowed to send to that provider | Regulated data, data-residency rules and networks without internet access |
|---|
GritRedact offers both: install Docker in your environment for fully offline processing with no network egress and local model inference, or use the sales-assisted SaaS. Either way it keeps a human review step before permanent removal and exports an audit record with input and output hashes. There is no public or self-serve API, SDK, MCP server or free tier; pricing is agreed with sales.
How to test PII redaction with known-value documents
You cannot measure what a tool missed unless you know what was there. Start with documents whose every sensitive value is recorded.
- Assemble known-value documents. Label a set of approved real documents, or start with fictional documents that ship with their values, such as the free KYC, banking and loan and diagnostics sample packs. Include clean and degraded copies of the same page. The packs are a testing resource, not a free tier.
- Run the redaction with the same policy you will use in production.
- Search the output for every known value. Extract the text, run OCR on the images and inspect the metadata, then look for each recorded value, including partial matches and reformatted numbers.
- Count misses and over-redaction per entity type and per condition (clean, degraded, handwritten). One blended figure hides where a tool fails.
- Keep sampling in production. If a reviewer finds zero misses in n randomly sampled pages, the rule of three gives an approximate 95% upper bound on the miss rate of 3 ÷ n.
The guide to evaluating redaction software covers the test set, sampling and the evidence to ask vendors for. To compare tools, see data masking tools and Presidio alternatives.
What PII redaction does not do
- Redaction is not authorization. If a person or system is not entitled to a record, block access instead of sending a redacted copy.
- Redaction alone is not HIPAA de-identification. HIPAA recognizes the Safe Harbor and Expert Determination methods, and the process must match the intended use and recipient.
- Redacted is not always anonymous. What remains, such as a job title, a rare condition or an address in a letterhead, can still identify someone.
What GritWorks does not replace
- GritWorks complements authorization, governance and e-discovery tools rather than replacing them. Access control, data catalogues, retention policy and legal review platforms keep their jobs; GritWorks prepares the documents, images and audio that move between them.
- Redaction is not authorization. If a person or system is not entitled to a record, block access instead of sending a redacted copy.
- Automated redaction alone is not a declaration of HIPAA de-identification. HIPAA recognizes the Safe Harbor and Expert Determination methods; policies, validation and human review must match the intended use and recipient.
- PII flagging is not redaction. GritTransform can flag potentially sensitive values in extracted results, but that does not remove them from the source. Use GritRedact when content must be removed.
- Engagement is sales-assisted. GritWorks has no public or self-serve API, SDK, MCP server or free tier, and publishes no price list: pricing is agreed with sales. Customer APIs are provided after onboarding.
Common questions
What is PII redaction?
PII redaction permanently removes personally identifiable information, such as names, account numbers, addresses, faces and voices, from documents, images and audio, so the recipient cannot see or recover it from the output.
What is the difference between PII redaction and PII masking?
Redaction removes values; masking hides or replaces them, for example with tokens or realistic synthetic values. Both should leave the original value unrecoverable from the shared file. PII flagging, which only marks values, is not redaction.
Is a black box over text in a PDF enough?
No, not if the box is only an annotation. The text underneath stays in the file. The text layer, the page image, any hidden OCR layer and the metadata must all be cleaned. Test by extracting text from the output and running OCR on it.
How do I test whether PII redaction worked?
Use documents whose sensitive values you already know. Run the redaction, then search the extracted text, the OCR of the output and the metadata for every known value. Count misses per entity type and per document condition.
Should PII redaction run on-premises or in the cloud?
It depends on where the unredacted file may go. A cloud service receives the original before anything is removed; an on-premises or air-gapped deployment keeps it inside your network. Regulated data and data-residency rules often point to on-premises.
Does PII redaction make data HIPAA compliant or de-identified?
Not by itself. Automated redaction alone is not a declaration of HIPAA de-identification, and redaction is not authorization. HIPAA recognizes the Safe Harbor and Expert Determination methods, and the process must match the intended use and recipient.