PII Detector

The pii_detector guardrail scans prompts for personally identifiable information (PII) like emails, phone numbers, and social security numbers.

Import

from pydantic_ai_guardrails.guardrails.input import pii_detector

Basic Usage

from pydantic_ai_guardrails import GuardedAgent
from pydantic_ai_guardrails.guardrails.input import pii_detector

guarded_agent = GuardedAgent(
    agent,
    input_guardrails=[
        pii_detector(),
    ],
)

Parameters

Parameter	Type	Default	Description
detect_types	list[str] | None	All types	PII types to detect
threshold	float	0.0	Minimum confidence threshold

Detected PII Types

Type	Pattern	Example
email	Email addresses	user@example.com
phone	Phone numbers	555-123-4567
ssn	Social Security Numbers	123-45-6789
credit_card	Credit card numbers	4111-1111-1111-1111
ip_address	IP addresses	192.168.1.1

Examples

Detect All PII

# Default: detect all PII types
guardrail = pii_detector()

Specific Types Only

# Only detect email and phone
guardrail = pii_detector(
    detect_types=['email', 'phone'],
)

HIPAA-Focused

# Healthcare-relevant PII
guardrail = pii_detector(
    detect_types=['ssn', 'phone', 'email'],
)

Violation Result

When triggered, returns:

{
    'tripwire_triggered': True,
    'message': 'PII detected in input: email, phone',
    'severity': 'high',
    'metadata': {
        'detected_types': ['email', 'phone'],
        'count': 2,
    },
}

Use Cases

• GDPR compliance: Block prompts containing EU personal data
• HIPAA compliance: Prevent PHI in healthcare applications
• Privacy protection: Keep user data out of LLM context
• Data minimization: Enforce data handling policies

Note

The built-in detector uses regex patterns. For more advanced detection with ML models, see llm-guard integration.

Caution

This guardrail detects PII but doesn’t redact it. If you need to process prompts containing PII, preprocess them before calling the agent.

Advanced: Custom PII Patterns

For organization-specific identifiers, create a custom guardrail:

import re
from pydantic_ai_guardrails import GuardrailResult, InputGuardrail

async def custom_pii_detector(prompt: str) -> GuardrailResult:
    patterns = {
        'employee_id': r'EMP-\d{6}',
        'account_number': r'ACC-[A-Z]{2}\d{8}',
    }

    found = []
    for pii_type, pattern in patterns.items():
        if re.search(pattern, prompt):
            found.append(pii_type)

    if found:
        return {
            'tripwire_triggered': True,
            'message': f'Custom PII detected: {found}',
            'severity': 'high',
        }
    return {'tripwire_triggered': False}

guardrail = InputGuardrail(custom_pii_detector)

Related

• Input Guardrails Guide
• llm-guard Integration
• Secret Redaction (for output)