Classify before you protect
The chapter starts with a question from Meridian's leadership: could the loyalty database be recovered if ransomware struck tonight? Backups were running and reporting success, but nobody had restored one in over a year. Aiko Tanaka's pinned reminder sums up the chapter: "A backup you have never restored is a hope, not a backup."
Because not all data deserves the same protection, the first step is a small classification scheme of three or four levels, such as Public, Internal, Confidential and Restricted. The labels should drive behaviour automatically: Confidential data is encrypted, restricted to those who need it, backed up to a stronger tier and watched by DLP. In Microsoft 365 a label can carry encryption with the file; in the cloud, a classification tag lets policy-as-code refuse unencrypted resources. Over-classifying everything as Confidential makes the label meaningless, so most data should be confidently marked Internal or Public. Start with the systems whose loss would hurt most, and use discovery tools such as Microsoft Purview to draft classifications for review.
Encryption in transit and at rest, and DLP in proportion
Encryption in transit means TLS 1.2 or higher with valid certificates, including on internal services, since an attacker already inside the network is the realistic threat. Encryption at rest protects against stolen laptops, unwiped disks and misconfigured storage; full-disk encryption should be enforced on endpoints, and sensitive cloud data can use customer-managed keys. Defining storage in infrastructure as code with HTTPS-only access, a TLS 1.2 minimum, private network access and versioning makes these settings hard to forget.
The book is clear about the limit: encryption does nothing against someone using valid credentials, because the system decrypts data for them as it would for any user. It does not replace access control, logging or backups. Field-level encryption in the application is reserved for the Restricted tier. DLP should be tied to classification, start with one or two recognisable data types, run in audit-and-notify mode before blocking, and be treated as a backstop for honest mistakes rather than a wall against determined insiders. It pairs with the detections built in Chapter 12.
The 3-2-1-1-0 rule
The older 3-2-1 rule (three copies, two media types, one off-site) still stands, but ransomware added two digits: one copy that is offline or immutable, and zero errors, meaning every backup is verified and restore-tested. Modern ransomware hunts backup servers, shares and backup credentials and destroys them before encrypting live data, because victims with good backups do not pay. The immutable copy preserves the choice not to pay; tested restores make that choice real.
At Meridian, the first restore drill of the loyalty database showed two days of transactions missing: log backups had silently stopped after a storage path changed, while full backups still reported success. After fixing log shipping and adding an alert, the second drill produced a complete database in forty minutes, which became the system's measured recovery time.
Immutable and air-gapped backups
Immutable storage, such as cloud object lock in compliance mode, refuses deletes and overwrites for everyone, including administrators and the account that wrote the data, until retention expires. Governance mode lets privileged users override it, so an attacker with those privileges could too. A surge of denied deletes is a strong ransomware signal worth alerting on. Air-gapped copies, such as tapes or rotated drives, are less convenient but cannot be encrypted while disconnected; many small teams combine both.
Retention should cover realistic attacker dwell time and any legal duty to keep records; legal hold is a separate mechanism for specific records. The identity that controls backups must be separate from everyday admin accounts, with its own MFA and separation of duties for deletion.
Testing restores
Restore tests get skipped because nothing visibly breaks when they are postponed, so the book asks for a scheduled, owned event with recorded results. A test restores a real system to an isolated environment, verifies integrity against a manifest, checks the data is usable and current, measures recovery time, and confirms the keys and credentials were available to the person doing it. Rotate the systems and the people running the drill; a failed test is a success of the process. An expert section adds application-consistent backups, point-in-time recovery, and restoring only onto clean infrastructure from backups that predate the compromise.
KEY POINTS
Key points
- Define a three- or four-level classification and apply it first to the most sensitive systems.
- Confirm full-disk encryption on every laptop and TLS 1.2 or higher on every service, internal ones included.
- Check backups against the 3-2-1-1-0 rule and close the biggest gap first, usually the immutable copy.
- Enable compliance-mode object lock for at least one backup copy of critical data.
- Give backups a separate least-privilege identity with its own MFA.
- Schedule a quarterly restore drill to an isolated target and record the measured recovery time.
QUESTIONS
Common questions
- What is the 3-2-1-1-0 backup rule?
- It extends the classic 3-2-1 rule. Keep three copies of the data on two different media types, with one copy off-site; add one copy that is offline or immutable so it cannot be altered or deleted; and aim for zero errors, meaning every backup is verified and restores are actually tested. The two added digits respond to ransomware that seeks out and destroys backups before encrypting live systems.
- Does encryption at rest protect against ransomware or stolen credentials?
- No. Encryption at rest protects data when someone obtains the storage itself, such as a stolen laptop, an unwiped disk or exposed storage viewed without credentials. An attacker who signs in as a legitimate user or uses a compromised application sees decrypted data just as a normal user would, and encrypted data can still be deleted or encrypted again. Access control, logging and backups are still required.
- What is the difference between compliance mode and governance mode for immutable backups?
- In compliance mode, object storage refuses every delete or overwrite during the retention period for everyone, including root and administrator accounts. Governance mode allows sufficiently privileged users to shorten retention or delete early, which means an attacker who steals those privileges can too. The Hardened Stack recommends compliance mode for the copy that must survive a full administrative compromise, with retention chosen carefully because it cannot be undone.
- How often should backup restores be tested, and what should a test include?
- The book recommends a quarterly restore drill on the calendar with a named owner, rotating which system is tested. Each drill restores a real system to an isolated environment, verifies integrity against a checksum or manifest, checks the data is complete and current, records the time to recover, and confirms that the person doing it had the needed keys and credentials. A job reporting success is not a test.
- Should backups use the same administrator account as the rest of the environment?
- No. If the same admin account that runs the environment can also delete backups, compromising that one account defeats the whole recovery strategy. Give backups their own least-privilege service identity, protect the backup console with separate strong MFA, and require separation of duties for deleting backup storage.