Your Backup Is a Rumor Until You Restore It
NOVEMBER 21, 2026 · UMERCS

Every ransomware conversation reaches the same sentence sooner or later: we have backups. It is the security equivalent of saying the spare key is under the mat. True, comforting, and useful to anyone who already knows where the mat is.
The uncomfortable part is that attackers reached the same conclusion years ago. The backups stopped being the thing that saves you and became the thing worth deleting first. If the recovery path disappears, the negotiation changes, and everybody in the room knows it.
The plan nobody ran
Sophos puts out an annual survey of organizations that lived through a ransomware attack, and the 2026 edition contains a genuinely encouraging number. Sixty-six percent of organizations whose data was encrypted restored it from backups, up twelve percentage points on the year before. Fifty-six percent of attacks succeeded in encrypting data at all, which tells you the other side of the ledger.
The gap between those two figures is where the real question lives. A backup that exists is an assertion. A backup that has been restored, into a working environment, at a known speed, from a known point in time, is a fact. Most organizations can produce the first one on request. Fewer can produce the second without scheduling something.
There is a second number in the same survey that gets quoted less. Backup-based recovery rose to 66 percent of encrypted-data cases, which means a third of organizations with encrypted data did not get there. Some had no usable backup. Some had one and could not reach it. The distinction matters less on the day than the outcome does: both end in a negotiation you did not want to be in.
Recovery speed is the number that gets skipped entirely. An organization that restores in four hours and one that restores in nine days both appear in the same statistic, and they are not in the same business. Downtime is where the money goes: payroll that cannot run, orders that cannot ship, a phone system that cannot take a call. The backup worked. The business still lost the week.
One password, two data centers
A joint CISA and FBI advisory in August 2026 documented the Gunra ransomware operation, and one confirmed case in it is worth reading twice. The attackers deleted backup and archived data at the organization's primary data center and at its disaster recovery site. A single set of stolen credentials was enough to reach both.
The organization had done the thing the checklist asks for. Two copies, two locations. What it had not done was separate the trust. Both sites accepted the same administrator identity, so compromising that identity compromised the architecture. Geography is not isolation. A second building with the same keys is a second copy of the same problem.
Disaster recovery sites were designed for a different adversary. The threat model was a fire, a flood, a failed generator, something that destroys a building but not the credentials in it. Ransomware inverts that assumption. The attacker does not need the building. They need the account, and the DR site was built to accept that account as legitimate because it has to, every day, for replication to work.
How the way back gets found
This is not improvisation. BlackMatter made backup destruction standard procedure in 2021, hitting NEW Cooperative in Iowa and Crystal Valley in Minnesota. In both cases the group used compromised administrator credentials to locate every backup data store and appliance on the network, then wiped or reformatted them before encrypting anything else. CISA, the FBI and the NSA documented the tactic, along with ransom demands ranging from $80,000 to $15 million.
Change Healthcare in February 2024 is the expensive version of the same lesson. Attackers entered through a remote access portal with no multi-factor authentication. The backups were not isolated well enough to restore operations quickly. UnitedHealth paid $22 million and still did not get its data back, with total recovery costs estimated at $1.6 billion.
More recent incidents show the same pattern with better tooling. The Akira airline attack exploited a backup software vulnerability that had carried a patch for more than a year. The patch existed. The backup server simply was not in anyone's patch cycle, because it had been classified as an appliance and appliances are somebody else's job.
The pattern across all of these is the same three steps in the same order: find the backup infrastructure, take the identity that administers it, delete the recovery points before encryption starts. It is a sequence, which is good news for defenders. A sequence has steps, and steps can be interrupted. The step that matters most is the second one, because a compromised backup administrator account is the hinge the whole attack turns on.
The quietest room in the building
There is a structural reason this keeps working. Backup infrastructure is the least watched system in most environments. Detection and alerting get pointed at production, where users are. The backup server runs minimal logging, lighter access control, and older software, and it is administered by a small group who are usually busy with restores.
That is the whole opportunity. An attacker who lands in production is being watched. An attacker who moves to the backup server is, in many environments, alone in a room with the door closed. The reconnaissance is quiet, the deletion is fast, and the first alert anyone sees is the one from the person who tried to open a file.
The resourcing side is not a mystery either. In a survey of more than 1,100 IT and security professionals, 77 percent said their cybersecurity investment is not keeping pace with the threats they face, and 65 percent of MSPs said their clients are underinvesting. Knowing what good looks like and being able to fund it are different problems, and most teams are stuck on the second one.
There is also an inventory problem underneath all of this. Many organizations cannot produce a current list of where their backups physically and logically live, which means they cannot say who can delete them. An attacker with administrator rights does not need to guess. They enumerate, they find the management console, and the console is the single place where every recovery point in the environment can be reached at once. That is a design property, not a bug, and it is exactly why the console deserves more protection than the data behind it.
What actually closes it
Immutable storage is the baseline now, not an advanced measure. Object lock and write-once copies mean an attacker holding administrator credentials still cannot alter or delete the recovery point. That single property changes the arithmetic of the attack, because the destruction step stops working.
Then separate the trust, not just the location. Backup infrastructure should not share credentials or network paths with production. Multi-factor authentication and role-based access for the backup platform itself add friction exactly where an attacker is trying to move fast. Patch backup software on the same schedule as everything else, and treat a backup appliance as a server with an operating system, because that is what it is.
Finally, test restores on a schedule and include an attack scenario in the test. Recovering a file share is not the same exercise as rebuilding a domain controller while somebody is encrypting your file servers. Organizations that discover their restore does not work during an actual incident have no time left to fix it.
The practical question to ask an IT team or an MSP this week is short: when was the last restore test, what was restored, and how long did it take. If the answer arrives quickly and specifically, you have a recovery plan. If it arrives as a paragraph about policy, you have a document.
Sources
- BleepingComputer, Ransomware has a new target. Is your backup ready?, October 7, 2026: https://www.bleepingcomputer.com/news/security/ransomware-has-a-new-target-is-your-backup-ready/
- Sophos, The State of Ransomware 2026: https://www.sophos.com/en-us/content/state-of-ransomware
- CISA and FBI, joint cybersecurity advisory on Gunra ransomware, August 2026 (as documented in the BleepingComputer report above)
- CISA, FBI and NSA, advisory on BlackMatter ransomware backup destruction tactics (as documented in the BleepingComputer report above)
- IBM, Cost of a Data Breach Report: https://www.ibm.com/reports/data-breach
