Data Protection

Data Protection

What It Does

Data protection detects and redacts sensitive content before requests reach a model or tool.

It covers:

• PII: email, phone, SSN, credit card
• PHI-style patterns: medical record, patient ID, diagnosis, medication
• Secrets: API keys, bearer tokens
• Optional ML/NLP PII signals through Presidio

When To Use It

Use this when you want to inspect a plain string, redact a prompt, or make policy decisions based on sensitive data.

Import

from policyaware import DataProtectionEngine

Main APIs

API	Type	What It Does
DataProtectionEngine()	class	Creates the built-in rules-based sensitive-data detector.
engine.inspect(text)	method	Detects sensitive data and returns a DataFindings result.
engine.redact(text)	method	Detects sensitive data and returns redacted text in redacted_text.
PresidioPIIClassifier	optional ML class	Adds Microsoft Presidio-based PII/NLP detection when installed.

DataFindings Result Fields

engine.inspect(text) and engine.redact(text) return a DataFindings object.

Field	Type	Meaning
contains_pii	bool	True when PII such as email, phone, SSN, or credit card is detected.
contains_phi	bool	True when PHI-style data such as medical record, patient ID, diagnosis, or medication is detected.
contains_secrets	bool	True when secrets such as API keys or bearer tokens are detected.
contains_sensitive	bool	True when any PII, PHI, or secret is detected.
categories	list[str]	Detected categories, such as email, phone, ssn, api_key, or diagnosis.
redactions	int	Number of detected matches.
redacted_text	str | None	Redacted text when using engine.redact(text).

String Inspection

from policyaware import DataProtectionEngine

text = "Email jane@example.com or call 212-555-7890."

engine = DataProtectionEngine()
findings = engine.inspect(text)

print(findings.contains_pii)
print(findings.contains_phi)
print(findings.contains_secrets)
print(findings.categories)

Expected:

True
False
False
['email', 'phone']

Redaction

redacted = engine.redact(text)
print(redacted.redacted_text)

Expected:

Email [REDACTED_EMAIL] or call [REDACTED_PHONE].

Policy Fields

Data findings are exposed to YAML:

rules:
  - name: deny_secrets
    effect: deny
    when:
      data.contains_secrets: true

  - name: redact_pii
    effect: transform
    action: redact
    when:
      data.contains_pii: true

Optional Presidio PII

Install:

pip install "policyaware[presidio]"

Example:

from policyaware import PresidioPIIClassifier

classifier = PresidioPIIClassifier(score_threshold=0.5)
assessment = classifier.classify(
    "Krishna Kishor lives at 120 Main Street and his phone is 212-555-7890."
)

print(assessment.model_dump())

Use Presidio when you need stronger detection of names, locations, and addresses.