Join our Newsletter — 33% off our NHI Course
Home› Guides› AWS CloudTrail Guide: Organisation Trails, Data Events and…
Guide Cloud Security

AWS CloudTrail Guide: Organisation Trails, Data Events and Lake

PDF for newsletter subscribers · free to read online

← All guides
By NHI Mgmt Group, edited by Lalit Choda Updated 8 November 2026 12 min read
On this page

Almost every AWS investigation starts with CloudTrail, and almost every one discovers a gap: the trail was single-Region, data events were never enabled, the logs sat in the same account the attacker controlled, or someone stopped logging and nobody noticed. CloudTrail records API activity across AWS, but what it records depends entirely on how trails, event selectors and storage are configured. This guide goes deeper than the multi-cloud Cloud Logging and Audit Trail Guide: it covers organisation trails, management, data, network activity and Insights events, CloudTrail Lake and its May 2026 availability change, protecting the trail itself, and how to read identities out of events. It is written for cloud security engineers, detection engineers and incident responders working in AWS Organizations.

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

  • Event history is not a log strategy. It holds 90 days of management events per Region and per account only; you need an organisation trail delivering to a separate log archive account.
  • Data events are off by default. S3 object reads and writes, Lambda invocations and DynamoDB item access are invisible unless you enable them with event selectors, and they cost extra.
  • Protect the trail as a privileged asset. Disabling or modifying cloud logs is ATT&CK T1685.002; deny trail changes with service control policies and alert on any that succeed.
  • CloudTrail Lake closed to new customers on 31 May 2026. Existing customers can keep using it; everyone else needs another query layer such as Athena over S3 or a SIEM.
  • Identity is in every event. The userIdentity block, access key ID and role session details let you trace which human or workload acted, but only if role sessions are named and tied to a source identity.

How identity and access work in CloudTrail

CloudTrail has two jobs from an identity perspective. It records the identity behind every API call, and it is itself an AWS service whose configuration is controlled by IAM. Both matter.

Event categories

CategoryWhat it recordsDefaultInvestigation value
Management eventsControl plane operations: IAM changes, instance launches, security group edits, trail changes, console sign-insShown in event history; on trails you choose read, write or both (one copy free)Privilege escalation, persistence, configuration tampering
Data eventsData plane operations such as S3 GetObject, PutObject, DeleteObject, Lambda Invoke, SNS PublishOff; extra charges; not in event historyData access and exfiltration, ransomware against S3 (T1530)
Network activity eventsAPI calls through VPC endpoints (PrivateLink) for about 90 services, including denied calls (VpceAccessDenied)Off; extra chargesCredentials from outside the organisation probing endpoints
Insights eventsUnusual API call rates (write management events) and API error rates against a learned baselineOff; extra chargesSpotting bursts such as mass resource creation or enumeration

Reading identity from an event

  • userIdentity.type and arn: whether the caller was an IAM user, an assumed role, the root user, an AWS service or a federated identity, and its ARN.
  • accessKeyId: keys beginning AKIA are long-term IAM user keys; ASIA keys are temporary session credentials. Long-term keys appearing from new IP addresses are a classic compromise signal. See the AWS Access Keys Security Guide.
  • sessionContext and sourceIdentity: for assumed roles, the session issuer (the role) and session name. Requiring a meaningful session name or source identity when people and pipelines assume roles is what lets you tie a role session back to a human or workload. See the AWS STS and AssumeRole Guide.
  • sourceIPAddress, userAgent, errorCode: where the call came from, which tool made it, and whether it was denied. Bursts of AccessDenied from one principal usually mean enumeration.

Admin roles and privileged access

Whoever can change a trail can blind you. ATT&CK's T1685.002 Disable or Modify Cloud Log describes adversaries disabling CloudTrail and CloudWatch integrations, removing SNS topics, turning off multi-Region logging, or disabling log validation and encryption.

  • Organisation trails: only the management account, or a delegated administrator it nominates, can create or change an organisation trail. Member accounts can see it but cannot delete it, stop logging, change event types or otherwise modify it. This is the single most important structural control.
  • Delegated administrator: run trail administration from a security tooling account rather than the management account, which should hold as few workloads and people as possible. See the AWS Multi-Account Security Reference Architecture Guide.
  • Restrict AWSCloudTrail_FullAccess: AWS warns that this managed policy can disable or reconfigure auditing and should be granted only to account administrators.
  • Service control policies: deny cloudtrail:StopLogging, cloudtrail:DeleteTrail, cloudtrail:UpdateTrail and cloudtrail:PutEventSelectors in all member accounts except for a named break-glass role. Organisation trails already resist member changes, but account-level trails and event selectors still need this. See the AWS Organizations and SCPs Guide.
  • Log archive access: by default only the management account can read an organisation trail's bucket. Grant investigators read-only access through a dedicated role in the log archive account, never write or delete.

Integrations, API tokens and service accounts

CloudTrail depends on several machine identities and resource policies. Each is a place where a misconfiguration silently stops delivery.

ComponentIdentity or policyWhat to check
Member accounts in an organisation trailService-linked role AWSServiceRoleForCloudTrailRemoved automatically if an account leaves the organisation; logs already written stay in the bucket
S3 log bucketBucket policy allowing CloudTrail to write, with aws:SourceArn conditionNo public access (Security Hub CloudTrail.6), least privilege, server access logging (CloudTrail.7)
KMS keyKey policy allowing CloudTrail to encrypt and investigators to decryptSSE-KMS encryption (CloudTrail.2); a key policy error stops delivery; see the AWS KMS Security Guide
SNS notificationsTopic policy with aws:SourceArn or aws:SourceAccountPrevents other accounts publishing to or abusing the topic
CloudWatch Logs deliveryIAM role CloudTrail assumes to write to a log groupIntegration enabled (CloudTrail.5) and metric filters for critical events
CloudTrail Lake channelsChannel and resource policy for PutAuditEventsOnly named partners or applications can write non-AWS events

CloudTrail still creates an organisation trail in member accounts even when resource validation fails, for example a wrong bucket policy or insufficient KMS permissions. Check get-trail-status for delivery errors after every change, and alert on delivery failures, not just on the trail existing.

Logging and detection

Choosing data events without bankrupting the budget

Data events are high volume. Use advanced event selectors to choose precisely: fields include eventCategory, resources.type (for example AWS::S3::Object, AWS::Lambda::Function, AWS::DynamoDB::Table), resources.ARN, eventName, readOnly and userIdentity.arn. Wildcards are not supported, and adding advanced selectors to a trail overwrites any basic selectors, so read the current configuration with get-event-selectors first.

ApproachCoverageCostUse when
Write data events on all S3 bucketsEvery PutObject, DeleteObject, CopyObjectModerateDefault for most organisations; catches ransomware and tampering
Read and write on sensitive buckets onlyFull object access history for named ARNsContainedRegulated or crown-jewel data
All data events everywhereCompleteHighSmall estates or short forensic windows
Exclude noisy principalsAll except named service roles via userIdentity.arn NotEqualsReducedHigh-throughput pipelines with known roles

Watch for logging loops: a trail that logs all S3 data events and delivers to a bucket it monitors generates events about its own deliveries. Deliver to a bucket in the log archive account.

A known ransomware pattern against S3 shows why write data events matter: attackers with stolen keys re-encrypt objects with SSE-C and set lifecycle rules to delete them. CloudTrail records only an HMAC of the SSE-C key, but object-level events show the burst of overwrites and management events show the lifecycle change.

Detections to build first

  • Logging tampering: StopLogging, DeleteTrail, UpdateTrail, PutEventSelectors, deletion of the log bucket or KMS key changes. GuardDuty raises Stealth:IAMUser/CloudTrailLoggingDisabled when a trail is deleted or updated or its bucket is deleted.
  • Credential misuse: GuardDuty's InstanceCredentialExfiltration.InsideAWS and .OutsideAWS findings, built on CloudTrail, flag EC2 role credentials used from elsewhere. See the Amazon GuardDuty Guide.
  • Persistence: new access keys, login profiles or role trust policy changes for existing identities (T1098.001 Additional Cloud Credentials).
  • Root activity: any root user API call or console sign-in.
  • Enumeration: spikes in AccessDenied errors or Insights API error rate events from one principal (T1580 Cloud Infrastructure Discovery).

Query layer after the Lake change

AWS's documentation states that CloudTrail Lake will no longer be open to new customers starting 31 May 2026, and that existing customers can continue to use it as normal. For existing users, Lake stores events in immutable event data stores with Trino SQL queries, retention up to 3,653 days (one-year extendable pricing) or 2,557 days (seven-year pricing), Athena federation and ingestion of non-AWS events. New adopters should query the S3 trail data with Athena or send it to a SIEM. Event history remains available to everyone for 90 days but searches only one account and one Region with one attribute filter.

Investigation workflow

  1. Anchor on the credential: start from the suspect access key ID or role session and list every event it produced, across all accounts and Regions.
  2. Pivot on source: group by sourceIPAddress and userAgent to separate attacker activity from the legitimate owner's.
  3. Check the trail itself: look for tampering events and gaps in delivery, and run validate-logs over the incident window before relying on the files as evidence.

Metrics

  • Percentage of accounts and enabled Regions covered by the organisation trail (target: 100%).
  • Percentage of S3 buckets with at least write data events logged, and of sensitive buckets with read events.
  • Trail delivery failures per month and time to fix.
  • Time from a StopLogging or DeleteTrail attempt to alert (target: minutes).
  • Security Hub CloudTrail control pass rate across accounts.

Hardening baseline

  1. Create the organisation trail: multi-Region, read and write management events, from the management account or a delegated administrator, delivering to a bucket in a dedicated log archive account. One trail per partition if you use more than one.
  2. Turn on integrity and encryption: enable log file validation, which writes an hourly, chained digest signed with SHA-256 with RSA and lets validate-logs prove whether files were changed or deleted. Use SSE-KMS with a customer managed key.
  3. Protect storage: S3 Block Public Access, a least-privilege bucket policy with aws:SourceArn, versioning and either MFA delete or retention controls. MFA delete cannot be combined with lifecycle rules, so choose deliberately; logs are kept indefinitely unless lifecycle rules say otherwise.
  4. Add data events in stages: pilot write events for S3 in one organisational unit, measure volume and cost for a month, then extend. Roll back by removing the selector; management events continue unaffected.
  5. Add network activity events for high-value services such as S3, KMS and Secrets Manager, starting with VpceAccessDenied only.
  6. Lock configuration: SCPs denying trail changes, AWS Config rules multi-region-cloud-trail-enabled, cloud-trail-encryption-enabled and cloud-trail-cloud-watch-logs-enabled, and Security Hub CloudTrail controls.
  7. Wire detection: CloudWatch Logs or EventBridge rules for tampering and root use, GuardDuty enabled organisation-wide, and a tested query path for responders.

Small organisations with a handful of accounts can follow AWS Control Tower defaults and add S3 write data events. Large organisations should budget data events per organisational unit, exclude known high-volume service roles by ARN, and give the SOC a query layer with retention aligned to their regulatory requirement. See the AWS Control Tower Guide.

Common misconfigurations

  • Relying on event history: 90 days, management events only, one account and Region per search. Incidents are often found later than 90 days.
  • Logs in the same account as the workload: an attacker with administrator rights deletes the bucket or the trail. Use a separate log archive account.
  • Single-Region trail: attackers launch resources in Regions you do not use. Console organisation trails are multi-Region; CLI-created ones may not be.
  • No data events: the investigation can prove the attacker had S3 permissions but not what they read or deleted.
  • Delivery silently broken: a KMS key policy or bucket policy change stops delivery while the trail still shows as configured.
  • Unnamed role sessions: every action appears as one shared role, so you cannot tell which engineer or pipeline acted.

Framework mapping

FrameworkReferenceCloudTrail control
NIST SP 800-53 Rev. 5AU-2, AU-12Management, data and network activity event selection
NIST SP 800-53 Rev. 5AU-9, AU-10Log archive account, SCPs, log file validation
NIST SP 800-53 Rev. 5AU-11, AU-6Retention lifecycle, query layer and review
NIST CSF 2.0PR.PS-04, DE.CM-09Logs generated and monitored
MITRE ATT&CKT1685.002, T1078.004, T1098.001, T1530Log tampering, cloud account misuse, added credentials, cloud storage access
AWS Security HubCloudTrail.1 to CloudTrail.7, CloudTrail.9, CloudTrail.10Multi-Region trail, encryption, validation, CloudWatch integration, bucket exposure

Practitioner checklist

  • Run one multi-Region organisation trail per partition, managed from a delegated administrator account.
  • Deliver logs to a dedicated log archive account with SSE-KMS, Block Public Access and least-privilege policies.
  • Enable log file validation and test validate-logs on a sample period.
  • Enable S3 write data events organisation-wide and read events on sensitive buckets.
  • Deny trail changes with SCPs except for a named break-glass role.
  • Alert on StopLogging, DeleteTrail, UpdateTrail, PutEventSelectors and delivery failures.
  • Require role session names or source identity for human and pipeline role use.
  • Confirm your query layer: CloudTrail Lake if already a customer, otherwise Athena or SIEM.
  • Track Security Hub CloudTrail control results across every account.

Standards and references

Related NHI Mgmt Group resources: Cloud Logging and Audit Trail Guide · Cloud Forensics Guide · Cloud Incident Response Guide · AWS Security Guide · Cloud Credential Compromise Playbook

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.