Guardrail: mask U.S. personal data (SSN, phone, card)

Guardrail that masks U.S. personal data in AI assistant replies: SSNs, +1 phone numbers, personal emails and card numbers. Rules, 12 tests, code.

Règles

Mask before sending any data that identifies or reaches a person:
- personal email addresses (john.smith@…);
- U.S. phone numbers, with or without +1;
- Social Security numbers (123-45-6789);
- payment card numbers (checked with the Luhn algorithm).
Let through: generic addresses (support@, info@…), company names, amounts, dates, order numbers.

Cas de test

Code

import re

PATTERNS = {
    "email": r"[\w.+-]+@[\w-]+\.[\w.-]+",
    "phone": r"(?:\+1[\s.-]?)?\(?\b[2-9]\d{2}\)?[\s.-]?\d{3}[\s.-]?\d{4}\b",
    "ssn": r"\b(?!000|666|9\d{2})\d{3}-(?!00)\d{2}-(?!0000)\d{4}\b",
    "card": r"\b(?:\d[ -]?){13,19}\b",
}
GENERIC = ("support@", "info@", "sales@", "hello@", "contact@")


def luhn(number: str) -> bool:
    digits = [int(c) for c in number if c.isdigit()][::-1]
    total = sum(d if i % 2 == 0 else (d * 2 - 9 if d * 2 > 9 else d * 2) for i, d in enumerate(digits))
    return 13 <= len(digits) <= 19 and total % 10 == 0


def mask(text: str) -> tuple[str, bool]:
    """Replace personal data with [<type> redacted]; return the text and True if anything was masked."""
    found = False
    for kind, pattern in PATTERNS.items():
        def repl(m):
            nonlocal found
            value = m.group(0)
            if kind == "email" and value.lower().startswith(GENERIC):
                return value
            if kind == "card" and not luhn(value):
                return value
            found = True
            return f"[{kind} redacted]"
        text = re.sub(pattern, repl, text)
    return text, found

Coût estimé

Tokens d'entrée par appel : ≈ 871

Coût d'entrée seulement, sans la réponse du modèle. Prix publics des fournisseurs (Mistral AI, DeepSeek, OpenAI, Anthropic), tarif standard en dollars converti en euros au taux BCE du 29 septembre 2026. Estimation : 1 token ≈ 3,6 caractères.

Identifiant

contextetech--mask-us-personal-data

Type

Garde-fous

Fichier

mask-us-personal-data.guardrail.json

Format

JSON (application/json)

Taille

1 928 octets (1,9 Ko)

Encodage

UTF-8

Tokens estimés

≈ 534

Langue

Anglais

Licence

MIT

Usage commercial

Autorisé

Données personnelles

Fictives (inventées)

Version

1.0

Publié le

2 octobre 2026 à 21:54

Mis à jour le

2 octobre 2026 à 21:54

Stockage

en base

Utilisations

0

J'aime

0

mask-us-personal-data.guardrail.json

{
  "format": "contextetech/guardrail/v1",
  "id": "contextetech/mask-us-personal-data",
  "description": "Guardrail that masks U.S. personal data in AI assistant replies: SSNs, +1 phone numbers, personal emails and card numbers. Rules, 12 tests, code.",
  "tags": [
    "pii",
    "ccpa",
    "security"
  ],
  "license": "MIT",
  "language": "en",
  "type": "pii",
  "action": "mask",
  "rules": "Mask before sending any data that identifies or reaches a person:\n- personal email addresses (john.smith@…);\n- U.S. phone numbers, with or without +1;\n- Social Security numbers (123-45-6789);\n- payment card numbers (checked with the Luhn algorithm).\nLet through: generic addresses (support@, info@…), company names, amounts, dates, order numbers.",
  "cases": [
    {
      "input": "You can reach John at john.smith@gmail.com for updates.",
      "expected": "block"
    },
    {
      "input": "Call me at (415) 555-0132 tomorrow.",
      "expected": "block"
    },
    {
      "input": "My cell is +1 212-555-0187.",
      "expected": "block"
    },
    {
      "input": "SSN on file: 219-09-9999.",
      "expected": "block"
    },
    {
      "input": "Charge it to 4111 1111 1111 1111, exp 08/28.",
      "expected": "block"
    },
    {
      "input": "Send the invoice to maria.lopez@acme-corp.com.",
      "expected": "block"
    },
    {
      "input": "Email support@contextetech.com with any question.",
      "expected": "allow"
    },
    {
      "input": "Your order #458712 ships tomorrow.",
      "expected": "allow"
    },
    {
      "input": "The total is $1,250.00 including sales tax.",
      "expected": "allow"
    },
    {
      "input": "Your appointment is on 10/12/2026 at 2 pm.",
      "expected": "allow"
    },
    {
      "input": "Acme Corp approved the quote.",
      "expected": "allow"
    },
    {
      "input": "Hi, how can I help you today?",
      "expected": "allow"
    }
  ]
}

Vérifier un texte (Python)

import re

PATTERNS = {
    "email": r"[\w.+-]+@[\w-]+\.[\w.-]+",
    "phone": r"(?:\+1[\s.-]?)?\(?\b[2-9]\d{2}\)?[\s.-]?\d{3}[\s.-]?\d{4}\b",
    "ssn": r"\b(?!000|666|9\d{2})\d{3}-(?!00)\d{2}-(?!0000)\d{4}\b",
    "card": r"\b(?:\d[ -]?){13,19}\b",
}
GENERIC = ("support@", "info@", "sales@", "hello@", "contact@")


def luhn(number: str) -> bool:
    digits = [int(c) for c in number if c.isdigit()][::-1]
    total = sum(d if i % 2 == 0 else (d * 2 - 9 if d * 2 > 9 else d * 2) for i, d in enumerate(digits))
    return 13 <= len(digits) <= 19 and total % 10 == 0


def mask(text: str) -> tuple[str, bool]:
    """Replace personal data with [<type> redacted]; return the text and True if anything was masked."""
    found = False
    for kind, pattern in PATTERNS.items():
        def repl(m):
            nonlocal found
            value = m.group(0)
            if kind == "email" and value.lower().startswith(GENERIC):
                return value
            if kind == "card" and not luhn(value):
                return value
            found = True
            return f"[{kind} redacted]"
        text = re.sub(pattern, repl, text)
    return text, found