Implementing AWS CloudTrail organization trails for multi-account audits
Australian organisations often run workloads across several AWS accounts, even when the business is modest. Sydney, Melbourne and Brisbane each host an AWS Region, while development, finance and operational workloads commonly sit in separate accounts. A central audit trail gives security and platform teams one evidence source without asking every account owner to maintain identical logging.
An AWS CloudTrail organization trail suits that architecture. Created from the AWS Organizations management account or a delegated administrator, it records eligible events from member accounts, including accounts added later. It also supports multi-Region delivery and global-service events, which matter for IAM, STS, CloudFront and other control-plane activity outside an Australian Region.
Centralised evidence supports controls aligned with the Australian Signals Directorate’s Essential Eight, OAIC privacy obligations where applicable, and APRA CPS 230 for regulated financial institutions. It does not replace AWS Config, GuardDuty or resource inventories: CloudTrail records API actions, not the full state of an environment. Those services answer different audit and detection questions.
Delivery is useful only when coverage, retention and access are tested. The implementation below creates a tamper-evident archive in a Sydney home Region, enables global and multi-Region events, then validates member-account activity and alerting. It also accommodates Australian managed-service teams and daylight-saving changes.
Prepare the organization and permissions
Begin by enabling all features in AWS Organizations and ensuring every intended account has a unique ID and named owner. Enable root MFA in the management account and decide whether to administer the trail there or through a delegated administrator. Delegated administration suits larger Australian enterprises and managed-service providers.
Give the trail administrator only the permissions needed for trails, buckets, keys and log destinations. Place the archive and KMS key in a tightly controlled account. Member accounts should not be able to change or delete the organization trail, and service control policies can block attempts to alter protected logging resources.
Define account purpose and data classification before choosing event selectors. Exclusions must be deliberate, so tag production, finance, development and sandbox accounts, then test policies in a sandbox organisation before an account joins the estate.
Create the central S3 and KMS archive
Choose a home Region and uniquely named S3 bucket. ap-southeast-2 in Sydney is a common Australian starting point; Melbourne (ap-southeast-4) and Brisbane (ap-southeast-3) are alternatives. Enable Block Public Access, versioning, server-side encryption and TLS-only access. Object Lock can provide write-once retention.
Create a customer-managed KMS key in the same Region and allow the CloudTrail service principal to generate data keys. Grant kms:Decrypt separately. The bucket policy must let CloudTrail read bucket ACLs and write only beneath the expected AWSLogs prefix; conditions covering the account, Region and trail prevent unrelated services from adding evidence.
Apply S3 lifecycle transitions and test restoration and decryption. CloudTrail payloads can contain identifiers or personal information, so keep the archive within approved access and monitoring boundaries.
Configure the organization trail
Run the creation command from the management account or delegated administrator in the chosen home Region. Use a globally unique bucket name and substitute the archive account and KMS key identifiers:
aws cloudtrail create-organization-trail \
--trail-name organisation-audit \
--s3-bucket-name acme-organisation-audit-ap-southeast-2 \
--include-global-service-events \
--is-multi-region-trail \
--enable-log-file-validation \
--kms-key-id arn:aws:kms:ap-southeast-2:111122223333:key/KEY-ID
Global events expose IAM changes, while the multi-Region setting captures each supported Region. Log validation creates digest files that can expose altered objects. For CloudWatch Logs, create the log group and an IAM role trusted by cloudtrail.amazonaws.com with logs:CreateLogStream and logs:PutLogEvents, then supply both ARNs to the trail.
Use advanced event selectors after estimating the data-event cost. This example includes management, data, insight and API activity categories, while resource filters can narrow high-volume services:
aws cloudtrail update-organization-trail \
--trail-name organisation-audit \
--advanced-event-selectors file://audit-selectors.json
[
{
"Name": "audit-events",
"Field": "$.eventCategory",
"Equals": ["Management", "Data", "Insight", "ApiActivity"]
}
]
Include S3, Lambda or DynamoDB data events only where object-level auditing has a defined business or compliance purpose.
Validate coverage and alerting
Check describe-trails, get-trail-status and get-event-selectors in the home Region. Generate a harmless read from a separate member account, allow for normal delivery delay, then inspect S3 or CloudWatch Logs. Repeat the test after adding an account. This verifies enrolment, the bucket policy, KMS permissions and global-event capture together.
Route metric filters to CloudWatch Alarms and EventBridge, then send actionable events to the security team’s workflow. Useful detections include root-user use, disabled MFA, guardrail changes, unauthorised Region use and repeated access-denied responses. CloudTrail can show who called the invalidation API, but it does not explain why a distribution remained stuck; the CloudFront invalidation troubleshooting guide helps connect API activity with delivery logs and cache state.
Keep source timestamps in UTC. Sydney and Melbourne change between AEST and AEDT, whereas Brisbane remains on AEST, so local-time dashboards can appear to jump twice a year. Convert only for display and calculate event-age alerts in UTC, avoiding false incidents during daylight-saving transitions.
Operate retention and audit evidence
Treat S3 as the long-term system of record rather than CloudTrail event history, which is intended for recent investigation. Apply the approved retention schedule, verify lifecycle transitions and periodically test restoration from archived storage. Use Object Lock where audit policy requires immutable evidence. Review KMS decrypt permissions and bucket policies quarterly and after staff or supplier changes.
Integrate selected events with the organisation’s SIEM while retaining the original S3 objects as the authoritative copy. CloudTrail shows that an API was called without showing the resulting configuration, so correlate it with AWS Config history, GuardDuty findings and change-management records. This gives Australian audit teams a request-to-effect sequence supporting Essential Eight, OAIC and APRA obligations that apply to the organisation.
Monitor CloudTrail, S3, KMS, CloudWatch and notification costs as selector scope grows. Data events, retained log groups and cross-account query traffic can become material. Assign an owner for selector changes, document the rationale in change tickets and retest after modifications. The finished trail becomes a durable, queryable and defensible audit record across the AWS Organization.