Suppose we have some user's personal data, that we have to encrypt and store in a database. I've read this would technically not be a breach of personal data, since the attacker would only have unintelligible data on his hands.

Is this true to your knowledge? Supposing the data would be appropriately encrypted, would an attack not count as a breach of personal data (something we would be held accountable for?)

1 Answer 1


Encryption is a mandatory safety measure which might mitigate some risks to data subjects and could therefore influence whether a breach must be reported. It does not affect your accountability or liability.

Per Art 32, the data controller is responsible for implementing appropriate technical and organizational safety measures. What is appropriate e.g. depends on the relevant risks, the state of the art, and the cost of implementation. Certain measures are explicitly suggested, and must therefore be considered. These include:

  • pseudonymization and encryption
  • measures to ensure the confidentiality, integrity, availability, and resilience of your services
  • ability to restore availability in case of an incident
  • regular testing and evaluation of your security measures

Certain approaches like full disk encryption and transport encryption are likely to be the bare minimum in this context, assuming state of the art encryption algorithms are used. Row-level encryption might provide stronger security guarantees, but isn't necessarily appropriate. Wherever encryption is appropriate it does not afford extra protection to the data controller – instead, a lack of appropriate encryption would be a GDPR violation.

When a data breach occurs, it is necessary to notify the supervisory authority and the data subjects.

The supervisory authority must be notified without delay, and at the latest within 72 hours of becoming aware of the breach. The notification is not required if “the personal data breach is unlikely to result in a risk to the rights and freedoms of natural persons”. It is possible to argue that strong encryption means that the risk is fairly low, and that the notification could therefore be skipped. However, this is a fairly risky argument to make and there are no downsides to making the notification. So even with encryption, it would be best to notify the data protection agency as a CYA measure. Regardless of whether the authority was notified, it is necessary to keep internal documentation about all such incidents.

Communication to data subjects is only required where there is a “high risk” to natural persons. Art 34(3) also lists certain conditions under which no individual notification of affected data subjects is necessary:

  • there have been appropriate technical and organizational measures, “in particular those that render the personal data unintelligible to any person who is not authorised to access it, such as encryption”; or
  • subsequent measures were taken so that the high risk can no longer materialize; or
  • individual notifications would require disproportionate effort, in which case a public notification should be made.

So it is quite possible that if you use appropriate encryption, then you might not be legally required to notify data subjects in case of a data breach.

Practical example: there is personal data on a work smartphone. The smartphone was lost on a train.

  • Scenario (a): the smartphone used full disk encryption so that the data is inaccessible unless a reasonably strong passphrase is entered. The phone contains measures against brute-force attacks and against physical data extraction. In that case, no real risk exists remains and affected data subjects don't have to be notified per Art 34(3)(a).
  • Scenario (b): the smartphone is enrolled in a mobile device management scheme that makes it possible to remotely erase all sensitive data. Once the remote erasure has been successfully carried out, the high risk has been mitigated and no notification to data subjects is necessary per Art 34(3)(b).

I find it more difficult to make such arguments about encryption for a database, because database servers are typically not breached directly. Instead, the application servers that connect to the database will be breached. Unless this is a backend for an E2EE application, these application servers will have access to the plaintext data. For example in the case of row-level encryption, the application server might hold the keys. But if this server is compromised, the keys would be compromised as well.

So before claiming that encryption would improve your GDPR compliance approach, it must first be determined whether such encryption is an appropriate measure in your specific context. In particular, I would expect care regarding the question how keys are managed so that they are shielded from attacks.

This answer assumes that a loss of control over encrypted data is still a personal data breach. I think that this interpretation is correct since encryption does not meet the GDPR's definition of anonymization, and since the GDPR repeatedly discusses encryption as a measure to protect personal data. This viewpoint also seems to be shared by the EDPB guidelines, such as their discussion of encryption as a supplemental measure to protect international transfers.

  • Thanks for the answer, I think I got a grasp of what's deemed "necessary" to make my application compliant. I've got one more question: let's suppose I make a hash of some sensitive data, and subsequently cut it by some characters (for other purposes, making it willingly un-decryptable). If such cut hash was to be "stolen" by some attacker, do you think this would still qualify as a data breach of sensitive data? Dec 14, 2021 at 15:22
  • Depending on context and on the amount of truncation, such a truncated hash might be directly identifying, might allow some indirect inferences, or might be truly anonymous. Note that WhatsApp's recent fine was in part for storing truncated hashes of phone numbers without a proper legal basis. On the other hand, I think that the k-anonymity model for the HIPB Pwned Passwords API uses truncated hashes in a reasonable and responsible manner. If you want a full answer about this, consider asking a separate question.
    – amon
    Dec 14, 2021 at 21:31

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .