- Subject
- Microsoft 365 & SharePoint
- Filed
- Loans
- 990
Cloud Backup Security: Air Gaps, Immutability and Recovery Planning
Separate backup access, retention and recovery paths to protect cloud data. Review sovereignty choices, deletion rules and restore readiness.

Cloud backup resilience depends on more than keeping another copy. IT teams need to know whether that copy is isolated from production, protected from changes, stored where required, and recoverable in the order the business needs. Those questions become harder as organizations spread workloads across Microsoft 365, Google Workspace, Salesforce, Azure and other cloud services.

Artificial intelligence adds another operational pressure. Copilots and agents can change content, permissions, configurations and workflows at a pace that manual recovery processes may struggle to track. One cited survey finding puts the issue in concrete terms: 62% of enterprises that felt confident in their defenses experienced AI-related unauthorized access in 2025. Confidence, by itself, did not prevent access.
The practical test for a backup program is whether it can produce a clean recovery point and return priority work to service. Encryption, geographic placement, retention and isolation matter because they support that outcome. They are not proof of resilience on their own; teams need to validate the controls and rehearse the restore path.
Cloud service availability does not cover every recovery need
Cloud providers secure and operate their platforms, but customers remain responsible for protecting their data, managing access, meeting their own compliance requirements and recovering from unwanted changes. A service cannot always distinguish an intended deletion from an accidental one, a safe script from a faulty one, or an approved configuration change from an unauthorized action.
That division of responsibility has business consequences. A lost system can interrupt revenue generating work and create downtime, compliance exposure and reputational harm. Those effects make backup planning a business risk decision as well as an IT task.
Replication and native recovery features can help, but they serve different purposes from an independent backup. Geo-redundant replication keeps a service copy available elsewhere; it may also replicate deleted, corrupted or encrypted data. Recycle bins and retention features can have time limits and may remain within the same administrative boundary as production.
Recovery planning also needs to cover each workload. Exchange, SharePoint, OneDrive, Teams and Entra ID each have distinct data and recovery needs. Teams and Entra ID can be overlooked when organizations focus mainly on files and email. For each service, confirm what is protected, how long recovery points remain available, who can change the settings, and whether restores work at the item, permission and metadata levels the business requires.
Protect the backup path with separate controls
A backup is only as trustworthy as the identities, storage and procedures around it. AvePoint describes Cloud Backup as encrypting data in transit with TLS 1.2/1.3 and using officially supported APIs that enforce encrypted connections. At rest, the company says each customer tenant uses a unique AvePoint-managed key, with a bring-your-own-key option for organizations that want to hold their own encryption key.
Isolation is another part of the design. AvePoint says backup data is segregated from production and held outside the Microsoft trust boundary under the customer’s selected policies. Organizations can choose a single region, configure multiple geographies or bring their own storage on Azure, AWS or Google Cloud. The company says data is not replicated across data centre regions and lists operations across North America, Europe and Asia-Pacific, including additions in Singapore and South Africa.

Role-based access can divide backup, restore, export and administrative duties. That separation matters because a backup account with broad privileges can become a route to the recovery copies it is meant to protect. Review the identity boundary, privilege level and approval process for each role. A separate storage location is less useful if production credentials can administer or delete it.
AvePoint says Cloud Backup does not require customers to deploy hardware, provision storage or maintain complex scripts. Its controls are independently audited against SOC 2 Type II and ISO 27001:2022. The company also says services meeting FedRAMP and CMMC Level 2 standards are available to US public sector and defence-industrial-base customers. These statements describe the vendor’s service and assurance claims; teams should check which service, region and workload are covered by the certifications relevant to their requirements.
Air gaps and immutability protect against different failures
An air gap limits how a compromise in production can reach a backup. Immutability prevents a backup from being altered or deleted during a defined retention period. One addresses access paths; the other constrains what can happen to data once stored. Using both can preserve recovery points when production controls fail, provided the identities and retention settings are configured correctly.
For cloud and SaaS data, a physical gap such as tape stored offsite may be difficult to operate. AvePoint describes its approach as a logical air gap: backup copies are segregated from the production tenant and governed by separate access controls. The goal is to prevent a compromised production environment from automatically granting access to the backups.
The service describes two immutability approaches. Policy-based immutability prevents unauthorized users from modifying or deleting backup data during a set period. Once that period ends, an authorized administrator can handle a valid deletion request, subject to applicable requirements. Storage-level immutability uses WORM, or write once, read many, storage. Under the described model, data is locked when written and cannot be changed by administrators or automated processes until the retention period expires.
Retention can also be set at a more granular level than one policy for an entire backup set. For example, a company might retain finance records for seven years and marketing data for two, according to the source’s example. The actual periods should follow each workload’s recovery needs and applicable retention obligations.
The 3-2-1-1 rule offers a useful design prompt: keep three copies of data, use two types of media, store one copy offsite, and keep one immutable or air-gapped. It does not replace testing. Short lock periods, poorly protected credentials and incorrect policy settings can still leave gaps. Monthly restore tests are the article’s recommended starting point, with frequency adjusted to workload importance and change.
A snapshot is not automatically an immutable backup. It captures a point in time, but it may still be possible for an administrator or attacker to remove it. Check the lock mechanism, retention enforcement, identity boundary and restore process before counting a snapshot as a protected recovery point.

Choose sovereignty controls to match the workload
Data sovereignty involves more than the physical location of a server. Organizations may need control over where information resides, who can access it, who holds the decryption keys, and how the provider could be compelled to provide access. The appropriate combination depends on the organization’s obligations and risk assessment.
Regional disruption, sanctions or outages can leave a company unable to reach its usual cloud environment. Recovery to a standby tenant, restoration across subscriptions, or migration out of a cloud can provide additional options. Organizations also assess how a service provider would handle demands for customer data or restrictions that affect staff access to technology. Those scenarios make independence of the recovery path a planning consideration.
AvePoint lists several options that change the level of customer control:
- Bring Your Own Key (BYOK): The organization controls the encryption key needed to decrypt its data.
- Bring Your Own Storage (BYOS): Backup copies reside in storage owned and managed by the customer.
- Regional residency and Multi-Geo support: Copies stay in a selected country or region. For supported Microsoft 365 or Google Workspace multi-geography configurations, AvePoint says it routes data to the corresponding data centre.
- Sovereign environments: Some organizations need locally operated services and personnel. The source points to UK Government Cloud, FedRAMP in the United States and IRAP in Australia as examples of different frameworks addressing local requirements.
- Citizen-operated infrastructure: Some customers require the people operating the environment to be citizens of a particular country.
- Data replicas: Additional geographically redundant copies add another layer of availability.
- Independent cloud deployment: Backup storage, processing and indexing can run on Azure or Google Cloud Platform independently of the production system.
- Repatriation: Organizations with that requirement can move away from public cloud infrastructure and operate independently.
More control can mean higher costs, more operational work or fewer service capabilities. A tightly restricted environment may also have less access to newer AI features or integrations. The right choice is the tier that fits the workload’s exposure, recovery objectives and operating capacity, rather than the most restrictive option available.
These controls can be combined. Regional residency addresses location; customer-held keys separate decryption authority; customer-owned storage changes custody and billing; and an independent cloud copy reduces dependence on the production provider. AvePoint says its Online Services let organizations select a home data centre and align it with Microsoft, Google or Salesforce regions. Teams should confirm the precise scope for each workload and region.
Set retention to protect recovery and privacy
Backup retention has to support recovery while respecting deletion obligations. The source discusses the tension between restoring data availability under the GDPR and the right to erasure under Article 17, alongside South Africa’s Protection of Personal Information Act (POPIA). It also notes that HIPAA and the Payment Card Industry Data Security Standard (PCI DSS) call for tamper resistant retention in relevant contexts.
A permanent lock can preserve data but make it difficult to meet a valid deletion requirement. A short retention period can reduce that conflict while leaving too few recovery points after an incident. There is no universal retention target: choose the shortest period that meets recovery, audit and business needs, then assess cost, deletion requests, legal holds and the time it could take to detect unwanted changes.
AvePoint says customers can set retention by workload or container, such as a mailbox, shared workspace or personal drive. For a verified erasure request, authorized administrators can approve deletion of the relevant data. The company says it logs who requested the deletion, who approved it and when it happened. The same approach must account for archived copies as well as backup copies, so a deletion completed in production does not leave the information elsewhere unnoticed.
For teams, the operational requirement is a documented path from request to decision and deletion. Identify who can approve the action, how the affected data is located across workloads, how the deletion is recorded, and what happens to held or otherwise restricted content. Backup and privacy processes need to agree on scope before a retention lock is set.
Prove that protected copies can restore priority work
Having a protected copy is only the start. Recovery depends on identifying a clean restore point, understanding dependencies and bringing essential operations back in a workable order. Executives need to know when people can resume work; a count of recovered terabytes does not answer that question.
AvePoint says Cloud Backup detects unusual activity so teams can investigate suspicious changes and select a trusted recovery point. The service can protect data up to four times per day, according to the source. It also supports restoring individual items, versions, permissions and metadata, either in place or to another location. Granular restores can limit the scope of a rollback when only part of an environment is affected.
AvePoint Cloud Backup Express integrates with Microsoft’s native Backup Service. The source describes Microsoft’s snapshot restores as reaching up to several terabytes per hour, combined with AvePoint’s recovery options in one interface. That speed claim is attributed to the service description; actual restore times depend on the workload and recovery conditions. Microsoft 365 Backup can add a high-speed recovery layer for supported workloads, while organizations may still need an independent copy for broader coverage, longer retention, granular restores, cross-tenant recovery, search, export or storage outside Microsoft’s trust boundary.
Recovery also needs a priority order. AvePoint’s Rapid Recovery System helps define a “Minimum Viable Company”: the smallest set of people, systems and data required for the organization to function. Teams can then restore those priorities first while other services follow. The useful preparation is to map dependencies, agree on what counts as essential, and rehearse the sequence before an outage forces decisions under pressure.
Review the design as workloads and automation change
Backup policies and restore tests should be revisited when workloads, business dependencies, recovery objectives or retention needs change. Critical services generally need more frequent validation than low change archives. The source suggests monthly testing for essential operations, quarterly testing for important workloads and an annual cross-functional recovery exercise as possible starting points; teams should adjust those intervals based on evidence.
For AI-driven changes, ask whether the backup service can flag unusual activity, preserve earlier versions, recover permissions and configuration, and identify which automated identity made a change. Check whether agent definitions and workflows need their own protection. Detection matters most when it leads to a clean restore point and a tested path back to service.
Air gaps, immutability, encryption, residency and retention each protect a different part of the recovery chain. Their value is proven through a successful restore that brings the right data and people back in the right order. Keep the design independent enough to withstand a production compromise, flexible enough to meet workload-specific obligations, and tested often enough that recovery is a practiced operation.

