
An encrypted archive can be perfectly confidential and completely unavailable to its owner. The storage service may be offline. A device may be lost. A separate key service may stop answering. None of these situations requires an attacker to break the cipher. They require a plan for which components must return, in which order, and with which credentials.
Follow the entire opening operation
Take a hypothetical architecture with an encrypted file in cloud storage and a separate service participating in key recovery. A backup of the file alone may not restore the ability to open it. A backup of a server alone may omit the client state required for the operation. “We have a copy” therefore needs a follow-up question: a copy of which dependency?
Inventory those dependencies before an outage. Include the file, necessary metadata, authorised client state, service configuration and any external identity dependency. The inventory should identify secrets without exposing them in the operational runbook. Otherwise the recovery document can become a new source of compromise.
Choose a recovery target, not just a backup schedule
NIST SP 800-34 Revision 1, the 2010 contingency-planning guide for federal information systems, is a useful starting point for connecting recovery priorities to business needs. It is not a certification requirement for every private messaging service.
The distinction between two targets helps a product discussion. Recovery time objective concerns the permitted recovery interval. Recovery point objective concerns the point in time to which data needs to be recovered. An organisation can restore service quickly and still lose work created after its last recoverable copy.
For a document service, the targets should cover more than object storage. A restored permission database that is older than the restored files can create a different problem from missing documents. The recovery exercise must examine whether the reconstructed state is consistent with the access policy the organisation intends to enforce.
The historical Safe workflow makes the question concrete
A UseCrypt Safe tutorial from 17 January 2019 demonstrates an optional setting to remove original files and directories after adding material. That option makes the location of the remaining usable copies an immediate operational question. The demonstration does not establish forensic secure erasure, nor does it show a recovery exercise.
Our historical HVKM analysis separately discusses the roles of client and server in recovering a file key. The appropriate conclusion is limited: the availability model must include those roles. A small or inexpensive server can still be a critical dependency. Its running cost does not measure the cost of its absence.
Test without weakening the privacy boundary
Begin in a controlled environment with representative non-sensitive files. Simulate loss of a client, unavailability of the participating service and restoration from an earlier backup. Measure actual time to open the intended documents. Check that revoked access does not return merely because a stale configuration was restored.
A successful exercise should not quietly introduce a universal decryption key that the published privacy model says does not exist. If recovery changes who can access content, that is a product and governance decision to disclose. Availability cannot be “fixed” by silently abandoning the access boundary people relied on.
Keep the result specific: which version was tested, which components failed, what was restored, what could not be recovered and how long each step took. A dated recovery test is more useful than an unlimited promise that encrypted files will always be accessible. Privacy and continuity belong in the same review because users need both, even though they are achieved by different controls.
