PII scrubbing
Third-party personal data masked in replies before they are sent.
Humane scans each outbound reply for personal identifiers. Phone numbers, emails, addresses and similar are masked in place with typed markers, while SSNs, card numbers and IBANs block the reply entirely.
PII scrub · outbound reply
Model wrote
Sure, you can reach Dana at 555-123-4567 or dana@example.com.
User receives
Sure, you can reach Dana at [REDACTED-PHONE] or [REDACTED-EMAIL].
phone, email, address, DOB, passport, licence
SSN, card number, IBAN
Mask what can be masked, block what can't
Rewriting a reply with a second model call is slow, costs tokens and can leak the same data in new words. Humane scrubs in place instead, so the reply keeps its shape and reviewers can see exactly what was removed. Where even a partial leak matters, the values boundary gate blocks the reply and sends a safe fallback.
- Masked: phone, email, postal address, date of birth, passport, driving licence
- Blocked: SSN, card numbers and IBAN
- Typed markers such as [REDACTED-PHONE] and [REDACTED-EMAIL]
- A pii_scrub summary on the response, without the original values
Gate stack · one outbound reply
- 1values_boundaryno policy line crossedPROCEED
- 2social_riskscore 0.42 · hold at 0.70PROCEED
- 3inaction_guardsubstantive, not hedgingPROCEED
- 4overclaim“you will definitely” · certaintyFLAG
- 5emotional_dependenceno permanence promisesPROCEED
- 6dissentopt-in, disabled by defaultOFF
See what was removed
r = client.process(user_id="user_42", message="what's Dana's number again?")
print(r.response)
# "You can reach Dana at [REDACTED-PHONE]."
for item in r.raw.get("pii_scrub", []):
print(item["category"], item["marker"])
# phone [REDACTED-PHONE]PII questions
Something else? Ask us.
Works well with
Safety gates
Before a reply reaches the user, Humane runs it through a fixed stack of gates and returns PROCEED, HOLD or BLOCK. Each gate records why it decided, and a gate that fails resolves to HOLD, never to PROCEED.
ExploreAudit trail
Humane records gate outcomes, safety flags, webhook deliveries, engine decisions and admin actions, each with its reason. When someone asks why the AI did something, you can show them.
ExploreTeams, SSO and roles
Each team's data is isolated at the query layer and gets its own memory store. Members sign in with email or Google, hold roles that control what they can change, and manage API keys that can be created, rotated and revoked.
ExploreEvery AI should know who it's talking to.
The safety and human-oversight layer for emotional AI: user state across sessions, risk alerts, safety gates and human takeover, with every decision explained.
