Zero-Knowledge Password Managers vs Vaultwarden: What “Can’t Read Your Vault” Still Leaks
Marisol Vega
September 18, 2026
The slogan is the same on the 1Password site, the Bitwarden trust page, and the Discord where someone just stood up Vaultwarden: we cannot read your vault. It is a true sentence about ciphertext. It is a false sentence about what still leaves the house. Email addresses, vault item counts, the domains you autofill, the IP that synced at 2 a.m., the browser extension that saw a password field on a lookalike site — none of that needs the AES key. People switch from Bitwarden Families to Vaultwarden to stop paying rent to a company, then treat the NUC as if it had signed the same slogan. The NUC can read more than they think, and the company could already see more than the slogan implied.
I ran Bitwarden for relatives until the bill felt like rent, then ran Vaultwarden until a power outage taught me which clients still unlock in airplane mode. This is not that outage story. This is what “can’t read your vault” still leaks while the box is up and the marketing is calm.
What zero-knowledge is allowed to mean
In the password-manager sense, zero-knowledge means the service stores a blob it cannot decrypt with the keys it holds in the ordinary course of business. Your master password (and, on 1Password, the secret key) never sits on their disk in a useful form. A warrant for “the vault” gets them ciphertext plus metadata. That is better than LastPass’s worst years. It is not a invisibility cloak.
Bitwarden official, 1Password, Proton Pass, and a well-built Vaultwarden all encrypt item contents. URLs, usernames, and notes are in the blob on current official clients — I am not writing 2015. What stays outside the blob, or sits beside it, is the leak surface: account email, organization structure, device list, last-sync timestamps, IP logs, support tickets you opened in plaintext, and anything the client phones home for “security reports.”
Vaultwarden is a Bitwarden-compatible server you run. It does not make the cryptography stronger than Bitwarden. It moves the admin role from a company with a policy to you with a volume. You can now read everything the server process can read. That includes more than ciphertext.

What a hosted zero-knowledge vendor still sees
1Password and Bitwarden will say they cannot open your logins. They can still see that you have an account, which family plan, how many items you have if their admin metrics count objects, which devices registered, and from which networks those devices checked in. Breach reports and “this password appeared in a dump” features need a way to talk about your domains or hashes. Watch what leaves the client when you enable those toggles. Some of that is a k-anonymity hash. Some of that is a list you did not mean to donate.
Support is the underrated leak. You paste a screenshot of a vault item into a ticket because the autofill is wrong. The slogan did not cover the PNG. I have done this for a relative. I regretted it.
Billing is another. The company knows your name, last four, and that you pay for six seats. That is not the vault. It is enough to make you interesting if someone is already interested in you.
Browser extensions see the page. That is the point of autofill. A malicious or compromised extension update, or a lookalike domain you trusted for one second, gets the password without anyone “reading the vault” on the server. Zero-knowledge at rest does not bind the DOM. 1Password and Bitwarden have both had to talk about extension security in public. The architecture requires the extension to be a privileged spectator.
Clipboard, screenshots, and OS password prompts are the same class. The vendor did not leak. Windows 11 and a screen-sharing cousin did. People still write the vendor a one-star review.
What Vaultwarden adds — and does not remove
On Vaultwarden you are the company. The Docker host sees the database file, the attachments directory, the proxy logs, and the fail2ban leftovers. Ciphertext in the SQLite or MariaDB is still ciphertext if you did not backdoor the image. The operator — you, a curious roommate, a ransomware that copies /data — gets:
- Every client’s IP and user-agent in reverse-proxy logs, unless you were careful.
- Account emails and organization membership in the clear tables that exist to make the API work.
- Vault size, attachment size, last revision times. A sudden 40 MB attachment is a story even if you cannot open it.
- The ability to ship a malicious Vaultwarden build, or a “debug” env, that starts logging more. Self-hosting is a supply chain you own. That is not automatically safer than Bitwarden’s.
- Backups of the volume on Backblaze or a USB disk. Those backups are the vault plus the metadata, sitting wherever you were sloppy.
What Vaultwarden does not add is immunity from the extension, the clipboard, or the phishing site. What it does not add is a secret key like 1Password’s unless you invent operational discipline that most NUCs do not have. What it does not add is “my ISP cannot see I use a password manager” if the HTTPS SNI still says vault.example.com.
I have seen people put Vaultwarden on a public hostname with a cute name and then argue they are more private than 1Password. They moved the metadata from a company that gets audited to a VPS that gets ignored. Different leak, same shape.

Metadata is the vault’s shadow
Even when item URLs live inside ciphertext, traffic analysis remains. Sync frequency tells you someone is working. A new device enrollment at 3 a.m. is a story. Emergency access and family sharing create social edges: who can request your vault, who is in the org. Those edges are not encrypted in the slogan’s sense. They are the product.
Send-a-password and share-a-item features create short-lived URLs. The vendor or your Vaultwarden sees that a share happened. The recipient’s network sees it too. Zero-knowledge did not cover the link.
Passkeys stored in the same vault inherit the same story. The private key is in the blob. The relying-party list and the sync chatter are still a shadow. I will not pretend a passkey in Vaultwarden is “more zero-knowledge” than a passkey in 1Password. The authenticator is the client. The server is still a mailbox for blobs.
The leaks that are your household, not the crypto
A spouse who knows the master password. A PIN on the browser that is four digits. Unlock with Windows Hello on a laptop that is also used for games. A printed recovery kit in a drawer a guest can open. Vaultwarden versus 1Password does not change any of that. I have seen more vaults “leaked” by a shared desktop session than by a server admin.
Airplane-mode unlock is availability, not secrecy. Offline clients hold a cached vault. That cache is a copy. Full-disk encryption and a lock screen are the controls. The slogan never mentioned them.
When I still pick a hosted zero-knowledge product
I pick 1Password or Bitwarden official when the users will not run a box, when I want someone else to get paged for an outage, and when I am willing to give that someone the metadata listed above. I pick 1Password when I want the secret-key ritual for relatives who will write it on a card once. I pick Bitwarden official when the price and the clients are enough and I do not want to be on-call for SQLite.
I do not pick them because the slogan made the leak surface zero. I pick them because the leak surface is a company’s and I can read the policy on a bored afternoon.
When I still pick Vaultwarden
I pick Vaultwarden when the bill is the problem and I will treat the host like a bank: updates, backups I have restored, admin on Tailscale not on the public internet, proxy logs rotated, image pin, no cute hostname. I tell relatives the unlock still works if my house is dark only if their clients already synced — that availability trade is what you lose when the Vaultwarden box goes dark, not what the ciphertext leaks.
I do not pick Vaultwarden to become more zero-knowledge than Bitwarden. I pick it to change who holds the metadata and who pays. If I am sloppy, I hold more leak surface than Bitwarden did, because I also hold the disk and the habit of SSHing in tired.
A comparison that stays honest
“Can’t read your vault” means they cannot decrypt item contents in the normal design. It does not mean they cannot count you, bill you, log your IP, register your devices, or watch your extension talk to a page. Vaultwarden does not erase that sentence. It replaces “they” with “I,” and it adds a volume you must not treat as junk.
If you need the contents hidden from a hoster, both designs can do that. If you need the fact of your vault hidden, neither can. Use a hostname that does not say vault, use a provider you accept, lock the clients, and stop quoting the slogan at the phishing site. I still run Vaultwarden for the cheap seats. I still tell people what it leaks. The icon was never the honest part.