Redact, mask, anonymise, or hash before any model sees the payload
FastPII gives engineering and security teams four explicit protection modes so every AI interaction can be transformed to the level your organisation actually needs.
Protection summary
The Protection Engine sits after sensitive data detection and before provider routing, giving you a clean place to decide which transformations should happen every time.
Choose the right transformation for the workflow
Some workloads need full removal, some need readable context, and some need one-way protection for downstream joins or evidence.
Redact
Replace the match with [REDACTED].
Use when nothing from the original value should cross the boundary to a model or downstream log.
Mask
Replace characters while preserving the original shape.
Useful when users need partial readability for support or review without exposing the full identifier.
Anonymise
Swap the value for a stable placeholder replacement.
Keep the semantic role of the data in the prompt while removing the original sensitive content itself.
Hash
Create a SHA256 one-way value.
Hashing is not reversible tokenisation. It is a one-way protection option for joins and traceability without raw exposure.
Use SHA256 hashing when you need one-way protection
Hash mode is designed for irreversible derivation. It is not reversible tokenisation, and it should be selected only when one-way behaviour is what your workflow requires.
Policy-friendly outputs
- • Redact for complete removal using [REDACTED].
- • Mask for character replacement with preserved shape.
- • Anonymise for placeholder replacement in prompts and logs.
- • Hash for SHA256 one-way protection.
curl https://api.fastpii.com/api/v1/protect \
-H "Authorization: Bearer $FASTPII_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"text": "Customer PESEL 90010112345 asked to update card 4242 4242 4242 4242",
"mode": "mask"
}'See how each mode transforms sensitive data
Every mode produces a different output shape. Pick the one that matches your workflow's readability and reversibility needs.
Before
Customer PESEL 90010112345 asked to update card 4242 4242 4242 4242
After
Customer [REDACTED] asked to update card [REDACTED]
Before
Customer PESEL 90010112345 asked to update card 4242 4242 4242 4242
After
Customer PESEL 9**********45 asked to update card 4*** **** **** *424
Before
Customer PESEL 90010112345 asked to update card 4242 4242 4242 4242
After
Customer PESEL [PESEL] asked to update card [CARD]
Before
Customer PESEL 90010112345 asked to update card 4242 4242 4242 4242
After
Customer PESEL a1b2c3... asked to update card d4e5f6...
Stop raw sensitive values from reaching downstream models
FastPII lets you decide how much context to preserve while still enforcing a consistent protection policy before AI traffic leaves your application.