Safety

Data redaction

Find personal and confidential values in prompts and replace them with placeholders before the model sees them.

Dozens of entity typesPlaceholders like [EMAIL_ADDRESS]Runs before the provider

What is Data redaction?

Data redaction is a guardrail that finds sensitive values in a prompt, such as email addresses, phone numbers and card numbers, and replaces each one with a typed placeholder like [EMAIL_ADDRESS] before the request reaches the model. Spro applies redaction as an input guardrail, so the provider never receives the original values. You choose which entity types to detect and the action to take.

People paste personal data into AI tools all the time, and once it reaches a model provider you have less control over where it goes. Redaction removes those values at the gateway, so the model still gets a useful prompt with the sensitive parts replaced.

Spro's sensitive information guardrail covers personal, financial, government and healthcare identifiers, including names, email addresses, card numbers, bank and tax identifiers, and national IDs for several countries. You choose the entities, and the redact action keeps the request going with placeholders.

Key features

What this feature does for your requests.

  • Detects personal, financial, government and regional identifiers
  • Replaces each match with a typed placeholder, such as [US_SSN]
  • Runs on the input before the provider receives the request
  • Choose entities one by one, so you only mask what you need
  • Pick redact, block, warn or log for each guardrail

How it works

4 steps, from setup to the first request.

  1. 01

    Enable the guardrail

    Open Guardrails, go to the Input tab and turn on sensitive information.

  2. 02

    Select entities

    Choose the types you need, such as email address and credit card number.

  3. 03

    Choose redact

    Matched values are replaced in place and the cleaned prompt is forwarded.

  4. 04

    Test with sample data

    Send a test prompt with fake values and check the result.

Is it the right fit?

Specific situations where this feature helps, and where it does not.

Good fit if

  • Support or healthcare chats where people paste contact or card details
  • Apps that handle local identifiers such as Aadhaar, PAN or US Social Security numbers
  • Teams that must keep personal data away from model providers

Not a good fit if

  • Free-text fields where identifiers are rare and hard to tell from ordinary numbers
  • A need for complete removal. Detection is pattern and context based and can miss values
  • Tasks where the model needs the real value to do its job

Data redaction FAQs

Common questions about Data redaction.

Route every AI request through SPRO.

One OpenAI-compatible API for 350+ models, with intelligent routing, 29 built-in guardrails, observability, and automatic failover.

Talk to sales