It's called a vault substitution attack, and it allows a malicious server to replace contents of a shared vault but also learn of any new items entered into that shared vault. The fix is pretty trivial from a cryptography perspective but it would require probably significant change in 1P applications and architecture/protocols.