Skip to content
Search
SecurityGlossary

PII Redaction

NIST SP 800-122, NIST's guide to protecting personally identifiable information, describes PII as information that can be used to distinguish or trace a person's identity, plus other information that is linked or linkable to them. That covers obvious fields such as names, email addresses and ID numbers, and also combinations, like a job title, a town and a date of birth, that identify someone when taken together.

Before you paste a support ticket, contract or spreadsheet into an AI tool, replace each identifier with a consistent placeholder such as [CUSTOMER_1] or [ACCOUNT_A], and keep the mapping on your side so you can put the real values back into the output. Pattern matching catches structured items such as emails, phone and card numbers. Names and addresses need a named-entity tool or a person's review. Redact logs and transcripts too, since they often hold more personal data than the prompt. Redaction works best alongside an approved tool and a clear AI acceptable use policy, and it's one of the most direct fixes for shadow AI risk.

Example

Original:
"Maria Lopez ([email protected], 555-0142) says card ending 4417 was charged twice on order 88213."

Redacted before sending:
"[CUSTOMER_1] ([EMAIL_1], [PHONE_1]) says card ending [CARD_1] was charged twice on order [ORDER_1]."

Prompt: Draft a reply to [CUSTOMER_1] about the duplicate charge. Keep the placeholders exactly as written.

(After the reply comes back, swap the placeholders for the real values on your own machine.)

When to use it

  • Pasting customer, patient or employee records into any AI tool
  • Building pipelines that send documents to a model API
  • Storing prompts, transcripts or logs

When not to use it

  • Assuming redaction makes any tool safe; data agreements and approved tools still matter
  • Leaving in combinations of details that identify someone even after names are removed