Skip to content

Building isolated, immutable backups for ransomware recovery on AWS

13 minute read
Content level: Advanced
0

This article details how to build an isolated, immutable backup foundation for ransomware recovery on AWS. It’s the second part of a 3-part series.

Introduction

Ransomware recovery depends on backups that stay isolated and immutable even when attackers reach your production accounts. AWS Support helps enterprise customers design and test these protected backup foundations with architecture reviews, runbook development, and ransomware recovery rehearsals. Technical Account Managers (TAMs), who lead these engagements, help confirm your recovery controls function before you require them.

This is Part 2 of a three-part series on designing a ransomware recovery architecture on AWS that aligns with the National Institute of Standards and Technology (NIST) Cybersecurity Framework (CSF). Part 1 established this risk and mapped the NIST CSF 2.0 outcomes the series addresses. This article builds the protected foundation. Part 3 covers how to test those backups and run a prioritized recovery.

Solution overview

This article builds the foundation in the following layers:

  • A dedicated data bunker account isolates recovery assets.

  • AWS Backup logically air-gapped vaults with Vault Lock make the recovery points immutable.

  • Multi-party approval adds human authorization for recovery.

Each layer maps to specific NIST CSF 2.0 outcomes.

Solution implementation

Prerequisites

Before you implement the architecture that this article describes, make sure that you have the following in place:

Building the data bunker account

The data bunker is a dedicated AWS account that’s separate from production workloads. The design limits IAM principals, avoids direct console access for day-to-day operations, and uses the account as a destination for cross-account backup copies.

The NIST Ransomware Profile connects this section to three outcomes. The text in quotation marks is the verbatim NIST CSF 2.0 outcome wording:

  • GV.RR-02: "Roles, responsibilities, and authorities related to cybersecurity risk management are established, communicated, understood, and enforced."\

    The account design and multi-party approval (MPA) team define who administers and authorizes access to the data bunker.

  • PR.AA-05: "Access permissions, entitlements, and authorizations are defined in a policy, managed, enforced, and reviewed, and incorporate the principles of least privilege and separation of duties."\

    Minimal IAM principals, no direct console access for day-to-day operations, and separated approval responsibilities apply these principles.

  • PR.IR-01: "Networks and environments are protected from unauthorized logical access and usage."\

    The dedicated account and network-isolation controls protect the environment that holds recovery assets.

Applying core design principles

To isolate the data bunker account, apply the following controls:

  • Grant only the minimal IAM principals, with strict access controls.

  • Block direct console access for day-to-day operations.

  • Restrict the account to use as a cross-account backup copy destination.

To configure the logically air-gapped vault, apply the following settings:

  • Configure your Vault Lock in compliance mode:\

    You can still change or remove the lock configuration during compliance mode’s minimum cooling-off period. After the cooling-off period expires, the lock prevents anyone from modifying or deleting backups, including the AWS account root user and AWS. For more information, see AWS Backup Vault Lock.

  • Use AWS owned encryption keys:\

    The default and recommended encryption for logically air-gapped vaults. Because AWS manages the key, you can’t delete or turn it off, that removes a dependency that a threat actor with account access can otherwise use. If your organization requires a customer managed key, then use one from an account in a separate recovery organization rather than the same account. For more information, see Encryption for backups in AWS Backup.

  • Set a Minimum retention period:\

    Enforced retention helps prevent premature deletion.

  • Set a Maximum retention period:\

    Maximum retention periods help maintain compliance with data governance policies.

The MPA team is a trusted, designated group that authorizes critical operations against the data bunker account, such as vault sharing during a recovery event. Include members from security, operations, and business leadership, and distribute them across geographies for coverage every day. Define the MPA team composition, escalation paths, and contact methods as part of your event response plan. For implementation details, see Multi-party approval for logically air-gapped vaults.

Support can review your data bunker account design, including the MPA team composition and the service control policy (SCP) and IAM guardrails that help protect the account, as part of an architecture review.

Isolating the network

The data bunker account operates with the following constraints:

  • No virtual private cloud (VPC) peering to production environments

  • No AWS Direct Connect or virtual private network (VPN) connections

  • Access only through AWS Backup service endpoints and the application programming interface (API)

  • AWS CloudTrail logging to a separate security account

Protecting the account

Although Vault Lock in compliance mode is designed to prevent the deletion or modification of recovery points, the data bunker account requires additional protection. The following controls help keep the account accessible during an event.

The NIST Ransomware Profile connects these account protections to three outcomes:

  • PR.AA-01: "Identities and credentials for authorized users, services, and hardware are managed by the organization."\

    Centralized root access and removal of root user credentials from the data bunker account help manage and control credentials.

  • PR.AA-05: "Access permissions, entitlements, and authorizations are defined in a policy, managed, enforced, and reviewed, and incorporate the principles of least privilege and separation of duties."\

    SCPs and IAM policies restrict account closure, account removal, and sensitive administrative actions.

  • GV.RR-02: "Roles, responsibilities, and authorities related to cybersecurity risk management are established, communicated, understood, and enforced."\

    Management account hardening establishes authority over the management and data bunker accounts.

Apply the following account-level controls:

  • Service control policies (SCPs):\

    Attach an SCP to the data bunker account’s organizational unit (OU) that denies the IAM action organizations:LeaveOrganization. Apply an IAM policy on the management account that denies account:CloseAccount, organizations:CloseAccount, and organizations:RemoveAccountFromOrganization for the data bunker account. For more information, see Essential security controls to prevent unauthorized account removal in AWS Organizations and Protecting member accounts from closure.

  • Centralize root access for member accounts:\

    Use IAM’s centralized root access management to remove root user credentials such as passwords, access keys, and signing certificates from the data bunker account. Removing root credentials eliminates the highest-privilege pathway a threat actor can use to bypass SCPs from within the account.

  • Management account hardening:\

    SCPs don’t apply to the management account. Protect the management account with multifactor authentication (MFA) on all principals, restricted access with no day-to-day use, temporary credentials only, and comprehensive CloudTrail monitoring.

  • MPA as a safety net:\

    If the data bunker account remains closed, then MPA provides an independent access path to the vault contents during the 90-day post-closure period.

Note: These controls help address unauthorized access to a single account within the same organization. If a threat actor compromises the management account, then you must use a cross-organization recovery pattern with an independent identity provider.

Defining the backup strategy

A comprehensive backup strategy for ransomware protection requires both formal planning and asset prioritization.

The NIST Ransomware Profile connects this strategy to three outcomes:

  • ID.IM-04: "Incident response plans and other cybersecurity plans that affect operations are established, communicated, maintained, and improved."\

    Your team communicates the documented strategy to stakeholders, reviews it regularly, and improves it through restore testing and event learnings.

  • ID.AM-05: "Assets are prioritized based on classification, criticality, resources, and impact on the mission."\

    The tiered strategy assigns backup protection according to asset criticality and mission impact.

  • PR.DS-11: "Backups of data are created, protected, maintained, and tested."\

    The strategy defines how you create and protect backups across tiers and maintain them through review and restore testing.

To establish and maintain the backup strategy as a documented cybersecurity plan, complete the following steps:

  1. Document the backup strategy as a cybersecurity plan.

  2. Communicate it to stakeholders and maintain it through quarterly reviews and regular tabletop exercises that simulate ransomware scenarios.

  3. Improve it based on restoration testing and event learnings.

  4. Integrate it into your event response procedures, with clear decision trees for when to restore or rebuild systems.

Stakeholders require training on their roles in prevention and recovery. This includes operations teams that run backups and business owners who understand RPO and RTO commitments.

Prioritize assets by business criticality across the following tiers:

  • Tier 0:\

    Critical assets such as customer databases, identity systems, and core applications. Frequent backups with logically air-gapped vaults, Vault Lock, and multi-party approval.

  • Tier 1:\

    Important assets such as supporting databases, configuration management, and logging infrastructure. Cross-account and cross-Region copies.

  • Tier 2:\

    Supporting assets such as internal tools, reporting systems, and non-critical integrations. Standard backup policies with moderate frequency.

  • Tier 3:\

    Non-critical assets such as development environments, archival systems, legacy applications, and static content. Standard backup policies with lower frequency.

This tiered approach improves both protection levels and cost efficiency and keeps recovery procedures clearly defined for each asset category.

Adding multi-party approval for human authorization

Even with logically air-gapped vaults, you require a mechanism to authorize recovery operations when a threat actor compromises the vault-owning account or it’s inaccessible.

The NIST Ransomware Profile connects MPA to three outcomes:

  • PR.AA-01: "Identities and credentials for authorized users, services, and hardware are managed by the organization."\

    The approval team uses managed AWS IAM Identity Center users, and only authorized individuals participate in the approval workflow.

  • PR.AA-05: "Access permissions, entitlements, and authorizations are defined in a policy, managed, enforced, and reviewed, and incorporate the principles of least privilege and separation of duties."\

    A multi-approver threshold and role separation help prevent one individual from authorizing access to protected backups.

  • GV.RR-02: "Roles, responsibilities, and authorities related to cybersecurity risk management are established, communicated, understood, and enforced."\

    Approval-team composition, escalation paths, and selection criteria define responsibility and authority for recovery authorization.

Understanding why multi-party approval matters

Multi-party approval (MPA) provides an independent authentication path. With MPA, multiple trusted individuals must authorize vault sharing through the approval portal before recovery can begin from a clean account. Your Technical Account Manager (TAM) can help you define the approval team, confirm out-of-band contact methods, and rehearse the approval workflow in a tabletop exercise. This rehearsal ensures approvers know their role before an event.

Manage your MPA approval team in a separate recovery organization that’s distinct from the primary organization that owns the logically air-gapped vaults. This separation helps prevent unauthorized access to the primary organization’s management account from also reaching the approval team. Then share the approval team with the vault-owning and recovery accounts with AWS Resource Access Manager (AWS RAM).

This article describes the approval team that lives in the primary organization’s management account. Treat this example as a starting point and evaluate a move to a separate recovery organization as your architecture matures. You can create and store MPA team resources only in the US East (N. Virginia) AWS Region. MPA commands and console operations must use US East (N. Virginia), regardless of where you locate your vaults.

Implementing the approval architecture

The following workflow shows how an approval team authorizes vault sharing from a clean recovery account:

  1. Approval team creation:
    An administrator creates an approval team in an Organizations management account. The team consists of trusted IAM Identity Center users. Then the administrator uses AWS RAM to share it with the vault-owning and recovery accounts.

  2. Vault association:
    AWS Backup administrators associate the approval team with new or existing logically air-gapped vaults.

  3. Recovery request:
    If a threat actor compromises the vault-owning account, then an initiator requests vault access from a clean recovery account and provides the vault Amazon Resource Name (ARN).

  4. Approval:
    The approval team reviews the request through the approval portal. When the minimum required number of approvers authorize the request, MPA automatically shares the vault with the requesting account.

  5. Recovery:
    After the approval team grants access, the recovery account can start restoring or copying data without waiting to remediate the affected account.

Use the following criteria to select approvers:

  • Geographic distribution, including different time zones for coverage every day

  • Role separation, including security, operations, and business leadership

  • Confirmed and tested out-of-band contact methods

  • Regular training and simulation exercises

Sharing vaults for recovery: AWS RAM and MPA

AWS Backup logically air-gapped vaults support two distinct mechanisms for sharing vault access with other accounts. Each serves a different purpose in your recovery strategy:

AWS RAM for proactive, pre-configured sharing:

Use AWS RAM to share a logically air-gapped vault with designated accounts, such as a forensics account or a recovery account, ahead of time. This mechanism supports planned operations such as restore testing or pre-staged recovery. After you share the vault through AWS RAM, the receiving account can directly restore backups from the vault without data transfer delays. For more information, see Logically air-gapped vault.

Multi-party approval (MPA) for emergency recovery when the vault-owning account is inaccessible:

Use MPA when a threat actor compromises the vault-owning account or locks it out. When an initiator submits a recovery request from a clean account, the approval team reviews and authorizes it through the approval portal. Then, MPA automatically shares the vault with the requesting recovery account. For more information, see Multi-party approval for logically air-gapped vaults.

Important: Configure both AWS RAM for planned recovery testing and MPA for worst-case unauthorized access scenarios. AWS RAM and MPA are independent integration paths. AWS RAM supports proactive vault sharing during normal operations. MPA supports emergency access when the vault-owning account is unavailable.

Conclusion

After you build the isolated data bunker account, the tiered backup strategy, and multi-party approval, complete the steps in Part 3 — Test backups and restore critical workloads after ransomware on AWS to validate your recovery points and restore your critical workloads in the order of their priority. To learn more about how Support can help, see AWS Support.

About the authors

Image

Francesc Sala is a Principal Technical Account Manager in the Strategic Industries team at Amazon Web Services. He has been 11 years in AWS helping enterprise customers to optimize cloud services as part of their journey to operational excellence. Prior to AWS, Francesc had 17-years experience in the Telecom and Networking industry.

Image

Luca Licheri is a Principal Technical Account Manager with 15+ years’ experience in infrastructures and cloud environments, working with global customers on their operational excellence. Luca is passionate about technology and enjoys reading and playing guitar in his free time.