Join our Newsletter — 33% off our NHI Course
Home› Guides› AWS Backup Security Guide: Vault Lock, Logically Air-Gapped…
Guide Cloud Security

AWS Backup Security Guide: Vault Lock, Logically Air-Gapped Vaults, Restore Testing and Backup Identities

PDF for newsletter subscribers · free to read online

← All guides
By NHI Mgmt Group, edited by Lalit Choda Updated 8 November 2026 13 min read
On this page

Ransomware operators who reach an AWS account with enough privilege delete or expire the backups first, so AWS Backup is only a recovery control if the vaults, keys and backup roles are protected from the same identities that run production. This guide covers AWS Backup specifically: Vault Lock in governance and compliance mode, logically air-gapped vaults and multi-party approval, vault access policies, the gaps where direct EC2 and RDS APIs bypass vault controls, restore testing, Backup Audit Manager and the CloudTrail events to watch. It builds on our vendor-neutral Backup Security and Immutable Backup Guide and is written for AWS platform, resilience and security teams.

This is an independent guide from NHI Mgmt Group, based on AWS's public documentation as of November 2026. It is not sponsored or reviewed by AWS. Product features change, so confirm settings against AWS's current documentation.

Key takeaways

  • Compliance mode is the immutable option. Once its grace time (minimum 3 days) expires, AWS states that no one, including the root user or AWS, can remove the lock, and recovery points cannot be deleted before their retention ends.
  • Vault controls only cover AWS Backup APIs. Direct EC2 calls such as ModifySnapshotAttribute and ModifySnapshotTier, and ec2:DisableImage on AMIs, are not governed by Vault Lock. Block them with IAM policies or SCPs.
  • Copy out of the blast radius. Logically air-gapped vaults are always compliance-locked, stored in a service-owned account and shareable to a recovery account; multi-party approval lets you restore even if the owning account is lost.
  • Keys decide recoverability. AWS recommends that a customer managed key for an air-gapped vault comes from a separate recovery account; keep backups and their key apart.
  • An untested backup is an assumption. Restore testing plans run scheduled restores, record restore times and clean up automatically, which turns RTO claims into evidence.

How identity and access work in AWS Backup

Four kinds of policy decide who can create, read, copy, change or delete backups, and an attacker only needs a gap in one.

LayerWhat it controlsNotes
IAM identity policiesWhich humans and roles can call backup:* APIsAWSBackupFullAccess includes deletion; avoid it for operators
Vault access policyResource policy on a vault, for example denying backup:DeleteRecoveryPoint to all but named principalsNo wildcards allowed in Action; cross-account requests are rejected for any action except backup:CopyIntoBackupVault
Vault LockRetention floor and ceiling, deletion and lifecycle changesGovernance (removable with IAM permission) or compliance (immutable after grace time)
KMS key policyWho can decrypt recovery points and copy themRestrictive key policies are a common cause of copy failures

Service roles. Backup and restore jobs run as IAM roles that trust backup.amazonaws.com. Restore testing selections need a role with the permissions in AWSBackupServiceRolePolicyForRestores, access to the key that encrypts the backups and to the key for the restored resource. AWS Backup also creates a service-linked role when you create a plan. These are non-human identities with read access to almost all production data, so scope them per purpose (backup, copy, restore test) and monitor their use. See the Backup Identity Isolation Guide.

Vault Lock: governance versus compliance mode

PropertyGovernance modeCompliance mode
Who can remove or change the lockUsers with sufficient IAM permissionsNo one after grace time, including root and AWS
Grace timeNot applicableChangeableForDays: 3 to 36,500 days
Vault deletionWith IAM permissionsOnly when empty
Recovery point deletionDenied while locked (lock removable)Denied until retention ends
CLIput-backup-vault-lock-configuration without --changeable-for-daysThe same command with --changeable-for-days
Use forTesting retention settings; workloads where an emergency override is acceptableRansomware resilience; records under SEC 17a-4, CFTC or FINRA (assessed by Cohasset Associates)

Retention limits: MinRetentionDays (minimum value 1) and MaxRetentionDays (up to 36,500) apply only to new backup and copy jobs; jobs outside the range fail, and existing recovery points keep their lifecycle. There is one lock per vault. Check status with DescribeBackupVault: Locked and LockDate (end of the grace period, UTC). Two warnings from AWS matter for design: recovery points with retention set to Always are kept forever once the grace time ends, and if the account is closed and not reopened within the 90-day suspension, AWS deletes the vault contents even if locked.

Choosing: start every new vault in governance mode, prove the backup plan's retention values fit within the lock range for a few cycles, then apply compliance mode with a grace time long enough to catch mistakes (for example 7 to 14 days). Small organisations with one account still gain most of the value from compliance mode on a single vault; large organisations should combine it with copies to a separate account.

Admin roles and privileged access

The question to ask is: can any single identity in the production account destroy every copy of a backup? If yes, the design has failed.

ActionRiskControl
backup:DeleteRecoveryPointDeletes backupsCompliance-mode lock; vault access policy deny except a named role
backup:UpdateRecoveryPointLifecycleShortens retention so backups expireVault Lock prevents lifecycle changes in locked vaults
backup:DeleteBackupVaultLockConfigurationRemoves a governance lock or a compliance lock in grace timeSCP deny for all but a break-glass role; alert on use
backup:PutBackupVaultAccessPolicy and backup:DeleteBackupVaultOpens or removes the vaultVault access policy deny; SCP
Changes to backup plans and selectionsSilently stops new backups for chosen resourcesManage plans through AWS Organizations backup policies; alert on changes
kms:ScheduleKeyDeletion and key policy changesMakes every recovery point unreadableKeys held in a separate account; SCP deny
ec2:ModifySnapshotAttribute, ec2:ModifySnapshotTier, ec2:DisableImageShares, re-tiers or disables backed-up snapshots and AMIs outside Vault LockIAM or SCP deny, exempting the backup plan roles
  • Separate administration: run organisation backup policies from a dedicated backup or security account. Production administrators should be able to request restores, not delete recovery points.
  • Break-glass: keep one tightly monitored role able to manage locks in governance mode and document when it may be used. See the Break-Glass Account Guide.
  • Human access: grant operators AWSBackupOperatorAccess-style permissions (start jobs, restore) rather than full access, and require just-in-time elevation for anything destructive.

Integrations, API tokens and service accounts

Logically air-gapped vaults and multi-party approval

A logically air-gapped vault is a secondary vault with Vault Lock always in compliance mode, encrypted with an AWS owned key (the default, which AWS recommends) or a customer managed key chosen only at creation. Backups are stored in an AWS Backup service-owned account, which CloudTrail shows as shared outside your organisation. Minimum retention must be at least 7 days. It can be shared through AWS Resource Access Manager with individual account IDs, including accounts in other organisations, but not with a whole organisation or OU, and backups can be restored from the account it is shared with.

Multi-party approval (an AWS Organizations capability) adds a recovery path that does not depend on the owning account. A recovery account in a separate recovery organisation requests access; an approval team of named individuals, sourced from IAM Identity Center in the recovery organisation, approves once a threshold set at team creation is met; the recovery account then gets a restore access backup vault that mirrors the source vault without storing data or being able to change it. Approval teams are created and stored only in US East (N. Virginia). There is no extra charge for the integration.

  • Keys: if you use a customer managed key, AWS recommends a key from a separate recovery account rather than the same account. For non fully managed resource types, the source backup must use a customer managed key to be copied, and that key must be shared with the copy role. Keeping the key away from the backups is what makes the encryption meaningful.
  • Limits: unencrypted Aurora, DocumentDB and Neptune clusters are not supported; several regions, including AWS GovCloud (US), do not offer the feature; if EC2 Allowed AMIs is enabled, add the alias aws-backup-vault to the allowlist or copies and restores fail.
  • Cross-region: AWS recommends cross-region copies into air-gapped vaults for greater resiliency.

Cross-account copies to standard vaults

Where air-gapped vaults are not available or suitable, copy to a standard vault in a dedicated backup account with a compliance-mode lock, a vault access policy and its own KMS key. Because cross-account requests are rejected for every action except backup:CopyIntoBackupVault, the production account can push copies but cannot delete them. Restrict that copy permission to accounts in your organisation with an aws:PrincipalOrgID condition. See the Cloud Native Backup and Ransomware Resilience Guide for the cross-cloud pattern.

Restore testing and audit evidence

Restore testing plans run on a cron schedule with a start window of 1 to 168 hours, choose the LATEST or a RANDOM recovery point from listed vaults within a SelectionWindowDays lookback, and restore up to one eligible recovery point per selected resource. Supported types include Aurora, DocumentDB, DynamoDB, EBS, EC2, EFS, FSx, Neptune, RDS and S3. Validation is optional and programmatic, with restored resources retained for 1 to 168 hours before AWS Backup deletes them. Restored resources are tagged awsbackup-restore-test (or named with that prefix for DynamoDB, S3 and others); removing the tag stops automatic deletion, so alert on that. AWS notes that cron schedules are evaluated within a day, so a plan meant to run every 12 hours with a start time after 11:59 runs only once a day.

Restore test resources are real copies of production data. Restore them into an isolated account or VPC with no inbound internet access, and restrict who can read them.

Backup Audit Manager evaluates active resources against controls in frameworks you create (for example, whether resources are backed up, encrypted, backed up daily, and whether they are in a plan with a Vault Lock) and publishes a report to S3 every 24 hours. It requires AWS Config resource tracking, which is charged. See the AWS Config Guide.

Logging and detection

SignalSourceWhy it matters
DeleteRecoveryPoint, UpdateRecoveryPointLifecycle API callsCloudTrailDeletion or retention tampering; attempts against locked vaults still show intent
PutBackupVaultLockConfiguration, DeleteBackupVaultLockConfiguration, PutBackupVaultAccessPolicy, DeleteBackupVaultCloudTrail API callsWeakening of vault protections
BackupDeleted, RecoveryPointCreated, BackupJobCompleted, CopyJobCompletedAWS Backup service events in CloudTrailMissing or failed copies; unexpected deletions
UpdateOrganizationalBackupPlan, DeleteOrganizationalBackupPlanOrganisation management events in member accountsChanges to centrally managed backup policies
AssociateBackupVaultMpaApprovalTeamFailed, CreateRestoreAccessBackupVaultFailedAWS Backup service eventsBroken recovery path through multi-party approval
kms:ScheduleKeyDeletion, kms:DisableKey on backup keysCloudTrailCryptographic destruction of backups
ec2:ModifySnapshotAttribute adding external accountsCloudTrailExfiltration of snapshots outside Vault Lock control

Send these to an alerting pipeline owned by security, not the backup team alone. GuardDuty and Security Hub help with the identity compromise that usually precedes backup destruction. See the AWS CloudTrail Guide and the Cloud Ransomware Guide.

Metrics:

  1. Percentage of critical resources with a recovery point in a compliance-locked vault outside the production account.
  2. Restore test success rate and measured restore time against RTO, per resource type, monthly.
  3. Backup Audit Manager compliance percentage for the frameworks in scope.
  4. Number of principals able to delete recovery points or remove locks, target a single break-glass role.
  5. Copy job failure rate to the backup account or air-gapped vault, with failures older than 24 hours.

Hardening baseline

  1. Centralise policy: enable AWS Backup in AWS Organizations, delegate administration to a backup account and deploy backup policies by OU so teams cannot opt out.
  2. Isolate: create a backup account (or a logically air-gapped vault) and copy critical backups there, cross-region where required.
  3. Lock: apply governance mode, verify plans run inside the retention range for two to four weeks, then convert to compliance mode with a 7 to 14 day grace time. Roll back during grace time with delete-backup-vault-lock-configuration; after it there is no rollback, so check values first.
  4. Close bypasses: SCP or IAM deny for ec2:ModifySnapshotAttribute, ec2:ModifySnapshotTier, ec2:DisableImage, lock deletion and KMS key deletion, exempting the backup roles. Test in a sandbox OU before rollout.
  5. Separate keys: keep customer managed keys for backup copies in the backup or recovery account.
  6. Multi-party approval: for the most critical data, set up a recovery organisation, IAM Identity Center and an approval team, and rehearse a restore through it.
  7. Test restores: create restore testing plans for each critical resource type, restoring into an isolated account, and feed results into resilience reporting.
  8. Audit: enable Backup Audit Manager with a framework that includes the Vault Lock control and review the daily report.

Framework mapping

ControlNIST SP 800-53 Rev 5NIST CSF 2.0CIS Controls v8.1ATT&CK risk addressed
Backup plans, copies, isolated vaultsCP-9 System Backup; CP-9(8) Cryptographic ProtectionPR.DS-1111.2 Perform Automated Backups; 11.4 Establish and Maintain an Isolated Instance of Recovery DataT1490 Inhibit System Recovery
Vault Lock, vault access policiesCP-9; AC-6 Least PrivilegePR.DS-11; PR.AA-0511.3 Protect Recovery DataT1485 Data Destruction
Restore testingCP-4 Contingency Plan Testing; CP-10 System Recovery and ReconstitutionRC.RP-0311.5 Test Data RecoveryT1486 Data Encrypted for Impact
CloudTrail alerts on backup tamperingAU-6 Audit Record Review, Analysis, and ReportingDE.CM-098.11 Conduct Audit Log ReviewsT1490

In ISO/IEC 27001:2022 the main control is A.8.13 Information backup, supported by A.5.30 ICT readiness for business continuity.

Common misconfigurations

  • Backups in the production account only: the same compromised administrator can delete workloads and backups, the pattern behind most cloud ransomware recovery failures.
  • Governance mode treated as immutable: any principal with the right IAM permission can remove a governance lock; attackers with administrator access can too.
  • Ignoring the EC2 side door: snapshots shared or AMIs disabled through EC2 APIs, which Vault Lock does not cover.
  • Keys next to the data: backup copies encrypted with a key administered from the production account, so deleting or disabling the key destroys every copy.
  • Retention set to Always in compliance mode: data kept forever and billed forever, which can also conflict with data protection deletion duties.
  • No restore evidence: backup job success reported as recovery readiness, with no scheduled restore tests or measured restore times.

Practitioner checklist

  • Manage backup plans through AWS Organizations backup policies from a delegated backup account.
  • Copy critical backups to a logically air-gapped vault or a compliance-locked vault in a separate account, cross-region where required.
  • Pilot every lock in governance mode, then move to compliance mode with a grace time of at least 7 days.
  • Deny recovery point deletion, lifecycle changes and lock removal to all but a monitored break-glass role.
  • Block direct EC2 snapshot sharing, re-tiering and AMI disabling outside the backup roles.
  • Hold customer managed keys for backup copies in the recovery account, never only in production.
  • Run restore testing plans for each critical resource type into an isolated account.
  • Alert on lock, vault policy, plan, key and snapshot attribute changes in CloudTrail.
  • Enable Backup Audit Manager with the Vault Lock control and review its daily report.

Standards and references

Related NHI Mgmt Group resources: Backup Security and Immutable Backup Guide · Backup Identity Isolation Guide · Cloud Ransomware Guide · Ransomware Response and Recovery Guide · Azure Backup and Recovery Services Vault Security Guide

PDF for subscribers

Take this guide with you

Every section, table and checklist in one PDF, with links to the sources and related guides. Subscribe to our weekly NHI & AI Identity Journal and we’ll email you a download link.

Free weekly newsletter

Subscribe to the NHI & AI Identity Journal

The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.

Bonus 33% off our NHI Course when you subscribe.

NHIMG Editorial Note
Researched and written by the NHI Mgmt Group editorial team using AI tools, under the editorial direction of Lalit Choda. Facts are drawn from the sources listed on this page and checked by our automated quality process. Last updated 8 November 2026. Spotted an error? Use our contact page.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM and Agentic AI security. nhimg.org

PDF Downloads for Newsletter Subscribers

Get the guide

Enter your details and we’ll email you a download link.

The full guide is free to read online. The PDF is a thank you for subscribing.

Fill in all fields marked * and tick the PDF box to continue.