Technology

At the point someone pulls off a valid MIM attack - which is basically a requirement here unless the whole BW/Vaultwarden server gets compromised- that is the least of someones problems. MIMs are incredibily hard these days.

Well if you decrypt the blob on the server they can see it.

That would be an understatement since all services claim your data is safe even in that case which is not true.

That's not very typical

BW06: Icon URL Item Decryption. Items can include a URL field, which is used to autofill the credentials and display an icon on the client. The client decrypts the URL and fetches the icon from the server, including in its request the domain and top-level domain of the URL. For instance, if the URL is “https://host.tld/path%E2%80%9D, the client request includes “host.tld”. This means that the adversary can learn (part of) the con- tents of URL fields. Using Attack BW05, an adversary can place the ciphertext of sensitive item fields, such as a user- name or a password, in the encrypted URL field. After fetch- ing the item, the client will then decrypt the ciphertext, confus- ing it for a URL. If the plaintext satisfies some conditions (i.e. containing a ‘.’ and no !), it will be leaked to the adversary. A URL checksum feature was deployed in July 2024, mak- ing the clients store a hash of the URL in another encrypted item field, therefore providing a rudimentary integrity check and preventing this attack. Note that old items are never up- dated to add such a checksum: this feature only protects items created after its introduction. Furthermore, URL checksums are only checked if a per-item key is present for the item. As we will see, an adversary can prevent per-item keys from being enabled with Attack BW10.

IMPACT. The adversary can recover selected target ciphertexts in the item, such as the username or the password.

REQUIREMENTS. The user opens a vault containing items that do not use per-item keys (i.e., items created before July 2024, or after Attack BW10 is run). The target plaintext must satisfy some additional conditions, detailed in Appendix

-- from the paper the article is discussing

So you could potentially expose your passwords to a compromised server or some kind of MITM. If they meet the conditions for the validation check, anyway.

And if the client software itself is compromised then all that is meaningless.