AWS Bedrock Delete GuardRails

SPLUNK detection rule by Bhavin Patel, Splunk · Apache-2.0

The following analytic identifies attempts to delete AWS Bedrock GuardRails, which are security controls designed to prevent harmful, biased, or inappropriate AI outputs. It leverages AWS CloudTrail logs to detect when a user or service calls the DeleteGuardrail API. This activity is significant as it may indicate an adversary attempting to remove safety guardrails after compromising credentials, potentially to enable harmful or malicious model outputs. Removing guardrails could allow attackers to extract sensitive information, generate offensive content, or bypass security controls designed to prevent prompt injection and other AI-specific attacks. If confirmed malicious, this could represent a deliberate attempt to manipulate model behavior for harmful purposes.

name: AWS Bedrock Delete GuardRails
id: 7a5e3d62-f743-11ee-9f6e-acde48001122
version: 7
creation_date: '2025-03-25'
modification_date: '2026-05-13'
author: Bhavin Patel, Splunk
status: production
type: TTP
description: The following analytic identifies attempts to delete AWS Bedrock GuardRails, which are security controls designed to prevent harmful, biased, or inappropriate AI outputs. It leverages AWS CloudTrail logs to detect when a user or service calls the DeleteGuardrail API. This activity is significant as it may indicate an adversary attempting to remove safety guardrails after compromising credentials, potentially to enable harmful or malicious model outputs. Removing guardrails could allow attackers to extract sensitive information, generate offensive content, or bypass security controls designed to prevent prompt injection and other AI-specific attacks. If confirmed malicious, this could represent a deliberate attempt to manipulate model behavior for harmful purposes.
data_source:
    - AWS CloudTrail DeleteGuardrail
search: >-
    `cloudtrail` eventSource=bedrock.amazonaws.com eventName=DeleteGuardrail |
    rename user_name as user | stats count min(_time) as firstTime max(_time) as
    lastTime values(requestParameters.guardrailIdentifier) as guardrailIds by src
    user user_agent vendor_account vendor_product dest signature vendor_region |
    `security_content_ctime(firstTime)` | `security_content_ctime(lastTime)` |
    `aws_bedrock_delete_guardrails_filter`
how_to_implement: The Splunk AWS Add-on is required to utilize this data. The search requires AWS CloudTrail logs with Bedrock service events enabled. You must install and configure the AWS App for Splunk (version 6.0.0 or later) and Splunk Add-on for AWS (version 5.1.0 or later) to collect CloudTrail logs from AWS. Ensure the CloudTrail is capturing Bedrock GuardRails management events.
known_false_positives: Legitimate administrators may delete GuardRails as part of normal operations, such as when replacing outdated guardrails with updated versions, cleaning up test resources, or consolidating security controls. Consider implementing an allowlist for expected administrators who regularly manage GuardRails configurations.
references:
    - https://docs.aws.amazon.com/bedrock/latest/userguide/guardrails.html
    - https://docs.aws.amazon.com/bedrock/latest/APIReference/API_DeleteGuardrail.html
    - https://attack.mitre.org/techniques/T1562/
drilldown_searches:
    - name: View the detection results for - "$user$"
      search: '%original_detection_search% | search user = "$user$"'
      earliest_offset: $info_min_time$
      latest_offset: $info_max_time$
    - name: View risk events for the last 7 days for - "$user$"
      search: '| from datamodel Risk.All_Risk | search normalized_risk_object IN ("$user$") | stats count min(_time) as firstTime max(_time) as lastTime values(search_name) as "Search Name" values(risk_message) as "Risk Message" values(analyticstories) as "Analytic Stories" values(annotations._all) as "Annotations" values(annotations.mitre_attack.mitre_tactic) as "ATT&CK Tactics" by normalized_risk_object | `security_content_ctime(firstTime)` | `security_content_ctime(lastTime)`'
      earliest_offset: 7d
      latest_offset: "0"
finding:
    title: User $user$ deleted AWS Bedrock GuardRails $guardrailIds$ from $src$
    entity:
        field: user
        type: user
        score: 50
threat_objects:
    - field: src
      type: ip_address
analytic_story:
    - AWS Bedrock Security
asset_type: AWS Account
mitre_attack_id:
    - T1685.002
product:
    - Splunk Enterprise
    - Splunk Enterprise Security
    - Splunk Cloud
category: cloud
security_domain: threat
tests:
    - name: True Positive Test
      attack_data:
        - data: https://media.githubusercontent.com/media/splunk/attack_data/master/datasets/attack_techniques/T1562.008/aws_bedrock_delete_guardrails/cloudtrail.json
          sourcetype: aws:cloudtrail
          source: aws_cloudtrail
      test_type: unit

[[ currentRule.title && currentRule.title.length > 60 ? currentRule.title.slice(0, 60) + '…' : currentRule.title ]]

ed2adf2f-6f8d-4eab-80a6-5a4e195e446b

Propose edits, view related rules, and engage with the community through comments.

Executable code, not a declarative rule

This submission is arbitrary [[ currentRule.format ? currentRule.format.toUpperCase() : '' ]] code, not a declarative detection rule — review it as you would any script before running it, and note that it cannot be safely aggregated or auto-executed alongside YARA/Sigma/Suricata/CRS rules.

Rejected pattern present

[[ reason ]]

What does this mean?

Format & documentation suggestions

Not a security issue — this rule is valid and was accepted as-is. These are optional fields/conventions its own format recommends, mainly affecting its quality score and how well it documents itself for other reviewers.

[[ reason ]]

What do these messages mean?

[[ currentRule.title && currentRule.title.length > 60 ? currentRule.title.slice(0, 60) + '…' : currentRule.title ]]

[[ currentRule.title ]]

Community Discussion