SOC 2 backup and disaster recovery for on-prem systems comes down to four things auditors check: backups that run and alert on failure, a copy stored far enough from the primary site that one event cannot destroy both, a tested restore that proves the recovery time objective (RTO), and a recovery plan the team has walked through.
These map to the Availability criteria A1.2 and A1.3 when Availability is in your report's scope, and to CC7.5 and CC9.1 in every report.
On-prem recovery differs from cloud region failover: physical media, offsite transfers, colocation provider coordination and manual restores. The criteria are the same for cloud and physical infrastructure, but the evidence looks different.
The AICPA points of focus that apply to backup and DR:
| Criterion | Point of focus | On-prem implementation |
| A1.2 | Determines Data Requiring Backup | Backup scope by tier |
| A1.2 | Performs Data Backup | Backup jobs, monitoring for failures, alerts and remediation tickets |
| A1.2 | Addresses Offsite Storage | A copy far enough from the primary site that one event cannot affect both |
| A1.2 | Considers Data Recoverability (names ransomware as a threat) | An out-of-band or immutable copy |
| A1.3 | Tests Integrity and Completeness of Backup Data | Quarterly restoration test |
| A1.3 | Implements Business Continuity Plan Testing | Annual tabletop exercise |
| CC7.5 | Implements Incident-Recovery Plan Testing | Tabletop scenarios such as ransomware or a lost site |
| CC9.1 | Considers Mitigation of Risks of Business Disruption | The disaster recovery plan and alternative processing |
A1.2 and A1.3 apply only when Availability is in the report's scope. Every SOC 2 report includes CC7.5 and CC9.1, and in our experience auditors still ask for backup and recovery evidence in a Security-only report.
Source: AICPA TSP Section 100, 2017 Trust Services Criteria with Revised Points of Focus (2022). Point of focus names are AICPA headings, and the implementations are Truvo's.
Recovery Point Objective (RPO)
The most data the organization accepts losing, measured in time. An RPO of 4 hours requires backups at least every 4 hours.
Recovery Time Objective (RTO)
The longest acceptable downtime after a disaster is declared. The RTO drives standby design, recovery staffing and coordination with the colocation provider.
Which mistake creates audit friction?
Targets set for how they sound instead of what the team can meet. An RPO of 1 hour fails when nightly full backups are the only mechanism and transaction log shipping is off: the real RPO is 24 hours. An RTO of 4 hours fails when restoring to a cold standby takes 8 hours including OS setup, deployment, restore and verification. Auditors accept realistic targets, and flag targets the evidence shows the team cannot meet.
Set RPO and RTO from tested recovery. Run a restoration test, measure the time, and set the RTO with margin for staff availability, troubleshooting and coordination with the colocation provider.
Each tier needs a different backup approach, using the same tiered asset classification as the rest of the on-prem SOC 2 cluster. A1.2 asks the entity to determine which data requires backup, and the tiers record that decision.
Tier 1: Critical Data
Production databases, application data, customer data, and any data whose loss would prevent the business from operating. Tier 1 gets the most frequent backups and the most thorough restoration testing.
Tier 2: System Configuration
Server configurations, application deployment configurations, firewall rules, network configurations, and infrastructure-as-code repositories. Configuration management or documentation can rebuild these, but backups make recovery much faster.
Tier 3: Operational Data
Logs, monitoring data, historical reports, and other data that supports operations but is not critical for business continuity. Backups can run less often, and Tier 3 loss does not trigger the DR plan.
Database backups come first. The approach depends on the platform:
| Platform | Backup Approach | Typical SOC 2 schedule |
| SQL Server | Native BACKUP DATABASE with full, differential, and transaction log types. SQL Server Agent handles scheduling and failure alerts. | Nightly full, hourly differential during business hours, transaction logs every 15-30 min for critical databases. |
| PostgreSQL | pg_dump for portable logical backups. pg_basebackup for physical backups with WAL archiving for point-in-time recovery. | WAL archiving gives near-continuous backup, and pg_dump gives a portable fallback restore to any PostgreSQL instance. |
| MySQL / MariaDB | mysqldump for logical backups. Percona XtraBackup for hot physical backups without table locks. | Percona XtraBackup is common in production because the tool avoids locking large databases during backup windows. |
All three platforms support scheduling through cron, built-in schedulers or backup management tools. Every backup job should alert on failure.
Backs up virtualized and physical on-prem environments: VM-level backups, bare metal recovery, application-aware backups and offsite replication. In our engagements, Veeam is the most common choice for VMware and Hyper-V. Application-aware backups put the database in a consistent state before the snapshot.
Backup and recovery across physical, virtual, cloud and SaaS workloads. In our experience, harder to deploy than Veeam, with broader workload coverage for large, mixed environments.
Open-source backup for complex topologies across mixed operating systems. No license cost, but more configuration work than commercial tools.
rsync and rclone
File-level backup for Linux. rsync handles local and remote sync, and rclone adds cloud storage targets (S3, Azure Blob, Google Cloud Storage) for offsite copies. For small teams, a scripted rsync job on cron is a light, auditable option: the cron job, transfer logs and offsite confirmation become the evidence.
Backup jobs often fail silently, and A1.2 expects monitoring for backup failures with corrective action. SNMP monitoring of backup storage (disk capacity, RAID status, controller health) catches hardware issues before backups fail. Feeding backup infrastructure into the same SIEM and monitoring stack used for security events shows both that the job ran and that the backup infrastructure is healthy.
What does A1.2 say about offsite storage?
A1.2 includes the point of focus Addresses Offsite Storage: backup data is stored far enough from the primary location that one security or environmental event is unlikely to affect both. On-prem, backups cannot live only in the same facility as production, because a fire, flood or long power outage at the primary colocation site would take out both.
The simplest approach replicates backups to a second colocation facility, cloud storage (S3, Azure Blob) or a geographically separate office with a backup server. The offsite location has to be independent enough that a disaster at the primary site does not reach the backups.
A second production site is a separate question from offsite backups, and the guide to SOC 2 Availability without a second data center shows how a single-site company meets A1.
With rclone or native cloud SDKs, the offsite transfer can run right after the local backup finishes, and the transfer confirmation becomes audit evidence.
How do out-of-band or immutable backups support recovery?
Ransomware that reaches the backup network can encrypt every copy the attacker can reach. We recommend a second copy that is out of band, on a separate network path with separate credentials, or immutable, with object lock that blocks deletion or changes for a set retention period. Air-gapped drives or tape are an optional extra safeguard. SOC 2 does not name immutable or air-gapped copies, but when Availability is in scope, A1.2 includes the point of focus Considers Data Recoverability, which names ransomware as a threat to plan for.
Options include cloud storage with object lock, which blocks deletion or changes for a set retention period, rotating external drives stored offsite, and tape, still used in regulated industries for this purpose.
The diagram below maps an on-prem backup setup: what each tier backs up, where the offsite and out-of-band or immutable copies live, and how the quarterly restore test proves the RTO.
Backups leave the facility twice: once to an offsite copy and once to an out-of-band or immutable copy. The quarterly restore test proves the RTO. Click to enlarge.
Backup and DR evidence follows the same three-part evidence model as other continuous controls, plus restoration tests and tabletop exercises.
What are the three parts?
A1.3 includes testing the integrity and completeness of backup data. A restoration test restores a backup to a non-production environment and confirms the data is usable. The evidence package includes:
We recommend at least one restoration test per quarter during the observation period. For a Type 1 report, one successful test shows the control is designed and implemented.
A1.3 calls for business continuity plan testing, and CC7.5 for incident-recovery plan testing, both built on scenarios. A tabletop exercise walks the team through a disaster scenario without failing over systems. The evidence package includes:
In our experience, one tabletop exercise a year is the minimum auditors expect. A 90-minute session walking through the primary colocation site is offline for 48 hours, what do we do? produces more audit value than a multi-day exercise that never gets scheduled.
In a colocation facility, backup and DR responsibility splits between the provider and your company.
What the colo provider handles
What the company is responsible for
Document the split in the subservice organization section of the audit package, and state in the backup and DR policy which recovery scenarios the colocation infrastructure handles and which need your own backup and restore. The bare metal SOC 2 readiness guide shows where this documentation fits.
Check backup job completion in alerts or the monitoring dashboard, usually 5 to 10 minutes, and investigate failures the same day. A failure left uninvestigated for more than 24 hours leaves a gap in the evidence.
Review backup storage capacity and health, confirm offsite transfers completed, check SNMP storage warnings, and confirm retention is enforced and old backups are pruned on schedule.
Restore the latest backup to a non-production environment, verify data integrity, record duration, scope and issues, and upload the evidence to the GRC platform.
Run a BCDR tabletop exercise, update the DR plan, compare RPO and RTO with tested results (if a restore took longer than the stated RTO, change the RTO), and update the backup and DR section of the Security Program Manual.
In small on-prem teams, the system administrator or infrastructure lead usually owns backup and DR.
Which roles are needed?
Our fractional security team builds effective security programs where backup and DR is a tested, audit-ready control.
Book Your Strategy CallLearn more about our SOC 2 for on-premise infrastructure services.
Further Reading: On-Prem SOC 2 Cluster
When Availability is in scope, A1.2 includes the point of focus Addresses Offsite Storage: backup data is kept far enough from the primary location that one security or environmental event is unlikely to affect both. Offsite copies in a second facility, cloud storage or rotated physical media meet that expectation. Auditors flag backups kept only in the same facility as production.
We recommend at least quarterly during a Type 2 observation period. For a Type 1 report, one successful restoration test shows the control is designed and implemented. Each test should restore to a non-production environment and verify data integrity. The test report, with duration, scope and results, is the evidence that the RTO is achievable.
A 90-minute meeting where the team walks through one disaster scenario: the primary site offline for 48 hours, ransomware encrypting production databases, or a critical server failing with no hot standby. The team agrees who does what, in what order and with what tools, and records gaps and action items. In our experience, one exercise a year is the minimum auditors expect.
SOC 2 does not explicitly require air-gapped backups, and air-gapped media is optional. In our experience, auditors increasingly ask about ransomware resilience, and A1.2 names ransomware under Considers Data Recoverability, so we recommend at least one copy that is out of band, on a separate network path with separate credentials, or immutable, with object lock that blocks deletion or changes. Rotated drives or tape can add an extra safeguard.
Document in the DR plan how the organization would recover from a ransomware event that encrypts network-accessible backup storage, and record the gap and an owner in the risk register until a separate or immutable copy exists.
The colocation provider covers facility redundancy (power, cooling, network) and physical security. Server-level backups, database recovery, application recovery and offsite storage are the company's responsibility. The provider's SOC 2 report covers physical security and facility availability, not backup, restoration or business continuity for the applications in the facility. Document that boundary in a shared responsibility model.
Yes, when four conditions hold:
Automated backups in AWS, Azure or Google Cloud then provide the backup evidence for A1.2 and the restore-testing evidence for A1.3 when Availability is in scope, and support CC7.5 and CC9.1 in every report. The cloud provider's SOC 2 report covers the storage infrastructure. Your team owns the backup configuration, access to the backup account and the restoration tests, and the auditor samples those.
The auditor reads the plan against CC7.5 and CC9.1, and with Availability in scope, A1.3. A plan that holds up names:
The evidence is the plan plus the test reports from the observation period.