How the attack works
Amazon GuardDuty analyses CloudTrail, VPC Flow Logs, DNS and other sources and raises findings for credential misuse, reconnaissance and known attack tools. Its detector is regional: DeleteDetector removes it with its findings, and UpdateDetector with enable false suspends it. In an organization, member accounts report to a GuardDuty administrator account; DisassociateFromAdministratorAccount (or the older DisassociateFromMasterAccount) in a member, or DisassociateMembers, DeleteMembers and StopMonitoringMembers in the administrator, cut that reporting. From the management account of the organization, DisableOrganizationAdminAccount removes the GuardDuty delegated administrator altogether.
Attackers who reach administrative rights disable detection early, usually in every Region they plan to use. Each call is a GuardDuty management event in CloudTrail (guardduty.amazonaws.com) with the caller in userIdentity and the detector ID or member accounts in requestParameters. Quieter variants such as trusted IP lists, suppression filters or disabling individual protection plans are not covered here; suppression rules and trusted and threat lists have their own page.
What an attacker needs
- Credentials for a principal allowed to call guardduty:DeleteDetector, UpdateDetector or the member and administrator disassociation actions.
Impact
No GuardDuty findings for the affected Regions or accounts, which blinds the SOC to credential abuse, crypto-mining and data exfiltration that would otherwise be reported.
Detection rules
GuardDuty Detector Deleted or Disabled
Detects DeleteDetector calls and UpdateDetector calls with enable false, which remove or suspend GuardDuty threat detection in a Region. It also matches attempts that failed (errorCode present): check errorCode, because an unauthorized attempt is worth seeing.
title: GuardDuty Detector Deleted or Disabled
id: 84bf96bd-f81d-4a87-8f3d-d68ca244adf8
status: experimental
description: >-
Detects DeleteDetector calls and UpdateDetector calls with enable false,
which remove or suspend GuardDuty threat detection in a Region.
It also matches attempts that failed (errorCode present): check errorCode,
because an unauthorized attempt is worth seeing.
references:
- https://docs.aws.amazon.com/guardduty/latest/APIReference/API_DeleteDetector.html
- https://docs.aws.amazon.com/guardduty/latest/APIReference/API_UpdateDetector.html
- https://docs.aws.amazon.com/eventbridge/latest/ref/events-ref-guardduty.html
author: Forestall
date: 2026-11-10
tags:
- attack.defense-impairment
- attack.t1685
logsource:
product: aws
service: cloudtrail
detection:
selection_delete:
eventSource: guardduty.amazonaws.com
eventName: DeleteDetector
selection_suspend:
eventSource: guardduty.amazonaws.com
eventName: UpdateDetector
requestParameters.enable: 'false'
condition: 1 of selection_*
falsepositives:
- Planned removal of GuardDuty from unused Regions; confirm the change record and prefer SCP-blocked Regions over deleted detectors.
level: high
False positives
- Planned removal of GuardDuty from unused Regions; confirm the change record and prefer SCP-blocked Regions over deleted detectors.
GuardDuty Account Detached from Administrator Monitoring
Detects GuardDuty calls that remove an account from central monitoring: a member disassociating from its administrator account, or the administrator disassociating, deleting or stopping monitoring of members. It also matches attempts that failed (errorCode present): check errorCode, because an unauthorized attempt is worth seeing.
title: GuardDuty Account Detached from Administrator Monitoring
id: 803547bb-1f5f-4eff-be94-8dc0d0ad903f
status: experimental
description: >-
Detects GuardDuty calls that remove an account from central monitoring: a
member disassociating from its administrator account, or the administrator
disassociating, deleting or stopping monitoring of members.
It also matches attempts that failed (errorCode present): check errorCode,
because an unauthorized attempt is worth seeing.
references:
- https://docs.aws.amazon.com/guardduty/latest/APIReference/API_Operations.html
- https://docs.aws.amazon.com/guardduty/latest/ug/logging-using-cloudtrail.html
author: Forestall
date: 2026-11-10
tags:
- attack.defense-impairment
- attack.t1685
logsource:
product: aws
service: cloudtrail
detection:
selection:
eventSource: guardduty.amazonaws.com
eventName:
- DisassociateFromAdministratorAccount
- DisassociateFromMasterAccount
- DisassociateMembers
- DeleteMembers
- StopMonitoringMembers
condition: selection
falsepositives:
- Accounts leaving the organization or being closed; confirm with the account inventory and the security team.
level: high
False positives
- Accounts leaving the organization or being closed; confirm with the account inventory and the security team.
GuardDuty Delegated Administrator of the Organization Removed
Detects DisableOrganizationAdminAccount calls to GuardDuty. The call removes the GuardDuty delegated administrator of the organization, and only the management account of the organization can make it. The delegated administrator account manages GuardDuty for the member accounts, so removing it takes central threat detection apart in one call. The Threat Technique Catalog for AWS lists the event with the other ways of disabling GuardDuty. The rule matches the event name: the affected account is the adminAccountId of the request. It also matches attempts that failed (errorCode present): check errorCode, because an unauthorized attempt is worth seeing.
title: GuardDuty Delegated Administrator of the Organization Removed
id: b9d4f630-e411-4046-94bb-10272eaea81a
status: experimental
description: >-
Detects DisableOrganizationAdminAccount calls to GuardDuty. The call removes
the GuardDuty delegated administrator of the organization, and only the
management account of the organization can make it. The delegated
administrator account manages GuardDuty for the member accounts, so
removing it takes central threat detection apart in one call. The Threat
Technique Catalog for AWS lists the event with the other ways of disabling
GuardDuty. The rule matches the event name: the affected account is the
adminAccountId of the request.
It also matches attempts that failed (errorCode present): check errorCode,
because an unauthorized attempt is worth seeing.
references:
- https://docs.aws.amazon.com/guardduty/latest/APIReference/API_DisableOrganizationAdminAccount.html
- https://docs.aws.amazon.com/guardduty/latest/ug/logging-using-cloudtrail.html
- https://aws-samples.github.io/threat-technique-catalog-for-aws/Techniques/T1562.A001.html
author: Forestall
date: 2026-11-18
tags:
- attack.defense-impairment
- attack.t1685
logsource:
product: aws
service: cloudtrail
detection:
selection:
eventSource: guardduty.amazonaws.com
eventName: DisableOrganizationAdminAccount
condition: selection
falsepositives:
- A planned move of the GuardDuty delegated administrator to another security account, which is a rare project with a change record. Confirm the record and that a new delegated administrator is registered.
level: high
False positives
- A planned move of the GuardDuty delegated administrator to another security account, which is a rare project with a change record. Confirm the record and that a new delegated administrator is registered.
Triage
- Identify the caller (userIdentity.arn), the Region (awsRegion) and the detector or member accounts in requestParameters, and check for an approved change.
- Look for the same principal making the same call in other Regions within a short time.
- Re-enable the detector or re-associate the account, then review the principal's activity in CloudTrail for the period without findings.
Response
Once the attack is confirmed:
- Turn detection back on in every Region where the call was made. Re-enable a suspended detector, create a new one where the detector was deleted, and bring detached accounts back under the GuardDuty administrator account.
- Contain the principal that made the calls by deactivating its access keys or revoking its role sessions, before it can switch the detector off again.
- Review CloudTrail for the period without findings, in every affected Region and account. A deleted detector takes its findings with it, so earlier findings are available only where they were exported.
- Look for quieter changes by the same principal that the rules do not match, such as trusted IP lists, suppression filters and protection plans that were switched off.
- Confirm that new findings from each restored Region and account arrive in the administrator account.
Mitigation
- Manage GuardDuty from a delegated administrator account with auto-enable for all members, so member accounts cannot leave on their own.
- Deny guardduty:DeleteDetector, UpdateDetector, DisassociateFromAdministratorAccount, DisassociateFromMasterAccount, DisassociateMembers, DeleteMembers and StopMonitoringMembers with an SCP except for a security break-glass role.
- Alert on detector state from AWS Config or Security Hub CSPM, not only on the API call.
Frequently asked questions
Are existing findings lost when GuardDuty is switched off?
It depends on the call. Suspending a detector (`UpdateDetector` with `enable` false) stops monitoring, and AWS documents that existing findings and settings remain intact. `DeleteDetector` removes the detector with its findings and configuration, and AWS documents that they cannot be recovered unless the findings were exported before.
Why do I have to check each Region separately?
A GuardDuty detector is regional. Deleting or suspending it affects one Region, so an attacker who wants no findings repeats the call in every Region they plan to use. Each call is its own CloudTrail event with the Region in `awsRegion`. The same principal making the same call in several Regions within a short time is a strong sign of intent.
Which ways of blinding GuardDuty do these rules not see?
Trusted IP lists, suppression filters and the disabling of individual protection plans leave the detector enabled, and the rules do not match them. The rules cover deleting or suspending the detector, cutting the link between a member account and the GuardDuty administrator account, and removing the delegated administrator of the organization. Review the quieter settings whenever one of the rules fires. Suppression rules and trusted and threat lists are covered on the page GuardDuty Suppression Rule or Trusted List Changed.
References
- MITRE ATT&CK T1685: Disable or Modify Tools
- Logging Amazon GuardDuty API calls with AWS CloudTrail
- Amazon GuardDuty API Reference: Actions
- Amazon GuardDuty events delivered to EventBridge (event source guardduty.amazonaws.com)
- Revoke IAM role temporary security credentials
- Suspending or disabling GuardDuty
- Amazon GuardDuty API Reference: DisableOrganizationAdminAccount