Security Insights · Security fundamentals · 8 min read
What Is Zero-Knowledge Encryption? How It Works and Its Limits
Zero-knowledge encryption explained: how client-side encryption keeps keys away from the provider, how it compares with other key models, and its limits.
Cloud services often say your files are encrypted, but that word alone doesn't tell you who can decrypt them. Zero-knowledge encryption answers that question directly: the provider that stores your files never holds the keys to open them.
This guide explains what zero-knowledge encryption means, how client-side encryption and key sharing make it work, how it compares with provider-held and customer-managed keys, and what it doesn't protect.
What is zero-knowledge encryption?
Zero-knowledge encryption means data is encrypted on the user's device before it is uploaded, using keys the service provider never holds. The provider stores and transmits only ciphertext, data in its encrypted form. Because it has no key, the provider can't read the files, and neither can an attacker who breaches its servers.
The term describes where the keys live, not a particular algorithm. Zero-knowledge services use standard ciphers such as AES. What changes is that encryption and decryption happen on users' devices, so the provider's servers never see anything but ciphertext.
How zero-knowledge encryption works
A zero-knowledge architecture rests on three steps. Details differ between products.
1. Client-side encryption
Each file is encrypted on the user's device before it is uploaded, usually with a random symmetric key generated for that file. Symmetric ciphers suit this job: NIST SP 800-175B notes that asymmetric algorithms tend to be much slower than symmetric ones, so they aren't used to process large amounts of data. The encrypted file goes to the server. The file key doesn't, at least not in readable form.
2. Keys that come from the user, not the provider
The file key needs protecting too. In a zero-knowledge design it is encrypted under keys that only the user controls: a key pair generated and kept on the user's device, or a key derived from the user's password. Password-derived keys need care. NIST SP 800-57 Part 1, NIST's key management guidance, explains that a password is less random than a cryptographic key, so the key-derivation function is run many times to make guessing slower. It requires the functions in NIST SP 800-132 when keys for stored data are derived from passwords.
3. Sharing by wrapping the file key for each recipient
To share a file, the sender's device encrypts the file key with each recipient's public key and uploads those small encrypted copies alongside the file. The file itself isn't encrypted again. Each recipient's device uses its private key to recover the file key, then decrypts the file locally. This is often called wrapping the key. NIST SP 800-57 calls the public-key version key transport and notes that the encrypted key can be stored for later decryption with the matching private key. The result pairs fast symmetric encryption for content with public-key cryptography for sharing, as our guide to symmetric vs asymmetric encryption explains.
Zero-knowledge vs provider-held and customer-managed keys
Cloud storage can protect stored files under three broad key models. What separates them is who holds the keys and where files are decrypted.
| Provider-held keys (encryption at rest) | Customer-managed keys (bring or hold your own key) | Zero-knowledge encryption | |
|---|---|---|---|
| Where files are encrypted | On the provider's servers, after upload | On the provider's servers, after upload | On the user's device, before upload |
| Who controls the keys | The provider | The customer, usually in a key service the provider's systems can call | Users and the people they share with |
| Where files are decrypted | The provider's servers | The provider's servers, using the customer's key | The recipient's device |
| Can the provider read files? | Yes | Yes, while the key is enabled | No |
| A breach of the provider's systems | Can expose readable files | Can expose readable files while the key is in use | Exposes ciphertext |
| Server-side search and previews | Available | Available | Limited, or done on the device |
| Lost user credentials | The provider restores access | The provider or customer restores access | Files may be unrecoverable without a recovery key |
Customer-managed keys give an organization control over a key's lifecycle, including the option to disable it and cut off access. In most designs, though, the provider's service still decrypts files on its servers to serve, index or preview them. Zero-knowledge encryption moves decryption to the user's device, so the provider has nothing to decrypt with.
What zero-knowledge encryption doesn't protect
Zero-knowledge encryption keeps file contents away from the provider. It doesn't secure everything else.
Metadata
A server has to know some things to store and route files. Depending on the product, it may see file names, file sizes, timestamps, account names and who shared with whom. Even NIST's definition of end-to-end encryption notes that routing information remains visible. Some services encrypt file names as well as contents. Sizes are harder to hide, because ciphertext is roughly the same length as the original file. If metadata matters to you, ask the provider which fields it can read.
A compromised device
Encryption on the device means decryption on the device too. If a laptop or phone is infected with malware, or someone uses an unlocked session, they can see files after they're decrypted. The same goes for a stolen password and second factor: whoever signs in as you can reach your keys. Device security, strong authentication and session timeouts still matter.
Lost credentials and account recovery
If keys are protected by a user's credentials, losing those credentials can mean losing the data. NIST SP 800-57 is direct about it: encrypted information can't be turned back into plaintext if the decryption key is lost. Services handle this with recovery keys, recovery codes or administrator recovery. Each one adds another path to the keys, and NIST frames key recovery as a trade-off between continuity of operations and the risk of exposing keys. If a provider can reset your password and restore your files without anything you hold, ask how, because that usually means it can reach the keys.
Is zero-knowledge encryption a zero-knowledge proof?
No. A zero-knowledge proof, as NIST defines it, is a cryptographic scheme in which a prover convinces a verifier that a statement is true without revealing any other information. Zero-knowledge encryption borrows the name loosely: it describes a storage or sharing service that has no knowledge of your keys or file contents. A product can use both, but one doesn't imply the other.
What to check in a zero-knowledge architecture
Zero-knowledge is a design claim, so it is worth testing. The Australian Signals Directorate's Information Security Manual (ISM) is a useful reference point. It approves one symmetric algorithm, AES, preferably AES-256, and calls for key management processes that cover key generation, distribution, storage, recovery and destruction. It also notes that encryption doesn't reduce the sensitivity or classification of data; it reduces the immediate consequences if malicious actors access it.
- Where encryption happens. On the device before upload, not on the server.
- Who holds keys. No key held by the provider, including escrow or backup copies, should decrypt files.
- How recovery works. Who holds any recovery key, and what a password reset does.
- Which algorithms are used. Established, published algorithms such as AES, not proprietary ones.
- Independent testing. Whether third parties test the security claims.
For questions specific to sharing, such as revocation and outside recipients, see our guide to end-to-end encrypted file sharing.
Where DekkoCORE fits
DekkoCORE is built on zero-knowledge, end-to-end encryption. Files are encrypted on the user's device with AES-256 keys before they're uploaded, and every file has its own key. DekkoCORE stores only ciphertext and never has the keys to open files, so not even DekkoSecure can read them. Files are decrypted on the recipient's device as they open them. Keys are handled automatically in the background: there are no HSMs, key servers or key ceremonies, and users never see a key.
Messages, eSignatures and video meetings are end-to-end encrypted the same way files are. Support sees a file only if a user chooses to share that one file with support, and that access is time-limited and logged. DekkoSecure uses established, publicly scrutinized cryptography rather than proprietary algorithms, and independent third parties test its security claims. Our guide to secure file sharing shows how this works day to day.
Common questions
Frequently Asked Questions
Is zero-knowledge encryption the same as end-to-end encryption?
They overlap but aren't identical. End-to-end encryption means data is decrypted only by its intended recipients, while zero-knowledge means the provider never holds the keys, including for stored files. A file sharing service can be both.
What is client-side encryption?
Client-side encryption is encryption performed on the user's device before data is sent to a server. It is the first step of zero-knowledge encryption, but it only counts as zero-knowledge if the provider also never receives the keys.
Can a zero-knowledge provider recover my files if I forget my password?
Not on its own, because it has no key to decrypt them with. Some services offer recovery keys or administrator recovery that you set up in advance. NIST SP 800-57 notes that encrypted information can't be turned back into plaintext if the decryption key is lost.
Is zero-knowledge cloud storage safe from hackers?
It protects stored files if the provider's servers are breached, because attackers find only ciphertext. It doesn't protect a compromised device or a stolen password and second factor, and metadata such as file names and sizes may still be visible.
Does zero-knowledge encryption hide file names and sizes?
Not always. Some services encrypt file names, but sizes, timestamps and sharing relationships are often visible to the server.
Which algorithms does zero-knowledge encryption use?
Usually AES for file contents and a public-key algorithm, such as elliptic curve cryptography or RSA, for sharing keys. The Australian ISM approves AES as its symmetric algorithm and prefers AES-256.