PII Detection Tools

from presidio_analyzer import AnalyzerEngine
from presidio_anonymizer import AnonymizerEngine

# Initialize engines
analyzer = AnalyzerEngine()
anonymizer = AnonymizerEngine()

# Detect PII
text = "John Smith's SSN is 123-45-6789 and email is john@example.com"
results = analyzer.analyze(
    text=text,
    language="en",
    entities=["PERSON", "EMAIL_ADDRESS", "US_SSN", "PHONE_NUMBER"]
)

# Print detections
for r in results:
    print(f"{r.entity_type}: {text[r.start:r.end]} (score: {r.score})")

# Anonymize
anonymized = anonymizer.anonymize(text=text, analyzer_results=results)
print(anonymized.text)
# "<PERSON>'s SSN is <US_SSN> and email is <EMAIL_ADDRESS>"

from presidio_analyzer import PatternRecognizer, Pattern

# Create a custom recognizer for employee IDs
emp_id_pattern = Pattern(
    name="employee_id",
    regex=r"\bEMP-\d{6}\b",
    score=0.9
)

emp_recognizer = PatternRecognizer(
    supported_entity="EMPLOYEE_ID",
    patterns=[emp_id_pattern]
)

# Add to analyzer
analyzer.registry.add_recognizer(emp_recognizer)

import spacy

# Load transformer-based model for best accuracy
nlp = spacy.load("en_core_web_trf")

# Add custom PII component
@spacy.Language.component("pii_detector")
def pii_detector(doc):
    pii_labels = {"PERSON", "ORG", "GPE", "DATE"}
    doc._.pii_entities = [
        ent for ent in doc.ents
        if ent.label_ in pii_labels
    ]
    return doc

# Register extension
from spacy.tokens import Doc
Doc.set_extension("pii_entities", default=[])
nlp.add_pipe("pii_detector", last=True)

doc = nlp("Dr. Maria Rodriguez from Mayo Clinic called on March 15.")
for ent in doc._.pii_entities:
    print(f"{ent.label_}: {ent.text}")

import boto3

comprehend = boto3.client("comprehend")

response = comprehend.detect_pii_entities(
    Text="Call John at 555-123-4567 or john@email.com",
    LanguageCode="en"
)

for entity in response["Entities"]:
    print(f"{entity['Type']}: score {entity['Score']:.2f}")

import google.cloud.dlp_v2

dlp = google.cloud.dlp_v2.DlpServiceClient()

inspect_config = {
    "info_types": [
        {"name": "PERSON_NAME"},
        {"name": "EMAIL_ADDRESS"},
        {"name": "PHONE_NUMBER"},
        {"name": "US_SOCIAL_SECURITY_NUMBER"},
    ],
    "min_likelihood": "LIKELY",
}

response = dlp.inspect_content(
    request={"parent": f"projects/{project_id}",
             "inspect_config": inspect_config,
             "item": {"value": text}}
)

from presidio_analyzer import AnalyzerEngine
from presidio_anonymizer import AnonymizerEngine

class LLMPIIGuardrail:
    def __init__(self):
        self.analyzer = AnalyzerEngine()
        self.anonymizer = AnonymizerEngine()

    def sanitize_input(self, prompt: str) -> str:
        """Strip PII from user prompt before sending to LLM."""
        results = self.analyzer.analyze(text=prompt, language="en")
        if results:
            anonymized = self.anonymizer.anonymize(text=prompt, analyzer_results=results)
            return anonymized.text
        return prompt

    def check_output(self, response: str) -> str:
        """Scan LLM response for leaked PII."""
        results = self.analyzer.analyze(text=response, language="en")
        if results:
            return self.anonymizer.anonymize(text=response, analyzer_results=results).text
        return response