The uncomfortable truth about ransomware is that the encryption event is not the start of the incident. By the time files are encrypted, an attacker has usually been inside the environment for days, learning where the backups live and how the recovery process works.
Businesses that recover quickly share a set of controls. Businesses that pay share a different set. The difference is rarely budget.
Why backups fail during a ransomware event
Most organisations have backups. Fewer have backups that survive an adversary who is actively looking for them.
The common failure modes:
- Backups are reachable with the same credentials as production. One compromised administrator account grants access to both the data and its copies.
- Snapshots are treated as backups. Snapshots in the same account, on the same platform, with the same permissions, are deleted first.
- Retention is too short. The compromise is discovered on day nine; the clean copy aged out on day seven.
- Restores have never been tested. The backup job reports success every night, and the first real restore reveals an exclusion, a missing key or a version mismatch.
- Recovery time is unknown. The team knows it has backups. Nobody knows whether restoring the core systems takes four hours or four days, which determines whether the business survives the week.
The baseline that holds
Immutable copies. At least one backup copy must be write-once for its retention period: object lock, an immutable appliance, or offline media. Immutability is a property of the storage, not a policy document.
Separate credentials and separate blast radius. The identity that runs backups should not be the identity that can delete them. Administration of the backup platform belongs in a different trust boundary from production administration.
Follow 3-2-1, then add the fourth. Three copies, two media types, one off-site. The fourth, which the original rule predates, is one copy that is offline or immutable.
Retention aligned to detection time. Assume a compromise runs undetected for at least thirty days. Retention shorter than that window may preserve only encrypted data.
Tested restores on a schedule. A quarterly restore of a real workload into an isolated environment, timed, recorded and reviewed. Anything less is an assumption.
An untested backup is a hypothesis. A tested restore is a control.
A written recovery sequence. Which systems come back first, in what order, with which dependencies, and who authorises each step. Recovery order is a business decision, not an infrastructure one.
Identity is the new perimeter
Most ransomware enters through credentials, not through a network vulnerability. The controls with the highest return are unglamorous:
- Multi-factor authentication on every administrative and email account, without exceptions for legacy protocols.
- Least privilege for service accounts, and a documented owner for each one.
- Removal of standing administrator rights in favour of time-bound elevation.
- Centralised logging that an attacker cannot edit, retained outside the production account.
- Alerting on the behaviours that precede encryption: mass permission changes, snapshot deletions, unusual authentication patterns.
What to decide before you need it
The decisions that determine the outcome are made in advance, in a room where nobody is under pressure:
- What is the maximum tolerable downtime for each system, expressed in hours?
- At what point does the business choose isolation and rebuild over restoration?
- Who has authority to make that call, including out of hours?
- What is the communication plan for customers, regulators and staff?
- Which cyber insurance obligations apply, and what evidence must be preserved?
Answering those five questions costs a few hours. Answering them during an active incident costs the business days it may not have.
A realistic standard
A mid-sized business does not need an enterprise security programme to be materially more resilient than its peers. It needs immutable backups with adequate retention, tested restores, strong identity controls, logging it can trust, and a recovery plan someone has actually read.
If you are unsure where you stand, start by asking one question: when did we last restore a production system from backup, and how long did it take? If the answer is not to hand, that is the first gap to close.