Encrypt the most sensitive fields

How to do it

Encrypt with authenticated encryption from a well-known library, such as AES-256-GCM or libsodium's secretbox. Use a fresh random nonce for every value and a key held in a secret store, not in the database. To search by the value, store an HMAC-SHA-256 of it under a second key. Disk encryption does not protect against someone who can query the database, as Encrypt database storage with a key you control explains. Store the key's version in front of each value. After a rotation, new values use the new key, and a background job re-encrypts old rows until no value names the old version. The rotation period lives in Keep encryption keys in a key service. Where a backup cannot be edited, encrypt each person's fields with a data key of their own. Destroying that key erases those fields in every copy, a method called crypto-shredding.

Examples

JavaScript example

import { createCipheriv, createHmac, randomBytes } from 'node:crypto';

function encryptField(plaintext, key, keyVersion) {
  const nonce = randomBytes(12);
  const cipher = createCipheriv('aes-256-gcm', key, nonce);
  const ciphertext = Buffer.concat([cipher.update(plaintext, 'utf8'), cipher.final()]);
  return `v${keyVersion}:` + Buffer.concat([nonce, cipher.getAuthTag(), ciphertext]).toString('base64');
}
const lookupHash = (value, hashKey) => createHmac('sha256', hashKey).update(value).digest('hex');

Python example

import base64, hashlib, hmac, os
from cryptography.hazmat.primitives.ciphers.aead import AESGCM

def encrypt_field(plaintext: str, key: bytes, key_version: int) -> str:
    nonce = os.urandom(12)
    sealed = nonce + AESGCM(key).encrypt(nonce, plaintext.encode(), None)
    return f"v{key_version}:" + base64.b64encode(sealed).decode()

def lookup_hash(value: str, hash_key: bytes) -> str:
    return hmac.new(hash_key, value.encode(), hashlib.sha256).hexdigest()

How to check

Query a sensitive column directly in the database. It shows unreadable ciphertext, and two rows with the same value hold different ciphertext. Thirty days after a key rotation, a count of values that name the old key version returns 0.

Sources

• GDPR Article 32 on security of processing
• OWASP Cryptographic Storage Cheat Sheet

Related rules

• DAT-80 Encrypt database storage with a key you control in Data and databases
• DAT-74 Give each tenant its own encryption key when customers ask to hold it in Data and databases
• ACC-81 Keep identity document scans apart and delete them after the check in Accounts and identity
• CMP-9 Mark special categories and children's data in Compliance and governance
• CMP-80 Use Swedish personal identity numbers only when clearly justified in Compliance and governance
• APS-54 Encrypt data on the wire and on disk in Application security
• APS-56 Use only standard cryptography in Application security
• APS-57 Keep encryption keys in a key service in Application security