An audit log tells you what happened, but only when someone reads it. If nobody is told when the root user signs in or when logging stops, you find out weeks later at the next audit, or halfway through an incident.
Send the events below to the channel your incident response team already watches, such as a Teams channel. See Do you set up Azure alert emails to go to a Teams channel?
| Event | Why it matters |
|---|---|
| A root user sign-in, in any account, successful or not | Root is break-glass access, so every use must be expected. See Do you lock down the AWS root user? |
StopLogging, DeleteTrail, UpdateTrail or PutEventSelectors in CloudTrail | Attackers turn logging off, or narrow it, before they do anything else |
PutBucketPolicy or DeleteBucketPolicy on the audit log bucket, and PutKeyPolicy, DisableKey or ScheduleKeyDeletion on its KMS key | Each can stop log delivery or lock you out of the logs |
| High-severity GuardDuty findings | Likely compromise. See Processing GuardDuty findings with EventBridge |
| No new audit logs for a set time, for example 1 hour | Logging broke, or someone disabled it in a way you did not catch |
Also alert on repeated failed sign-ins. See Do you monitor failed login attempts?
CloudTrail sends console sign-in events to Amazon EventBridge. Create a rule on the default event bus with this pattern, and send the matches to an SNS topic:
{"source": ["aws.signin"],"detail": {"userIdentity": {"type": ["Root"]}}}
The pattern leaves out detail-type on purpose. AWS spells it two ways, "Sign In" and "Signin", and a rule with the wrong one matches nothing and raises no error.
Two things catch people out:
For the CloudTrail, S3 and KMS calls in the table, write a second rule that matches "detail-type": ["AWS API Call via CloudTrail"] and the event names. See AWS service events delivered via AWS CloudTrail.
To get from SNS into Teams or Slack, use Amazon Q Developer in chat applications, formerly AWS Chatbot.
An alert on StopLogging misses quiet failures. A broken bucket policy or a disabled KMS key stops delivery without anyone calling StopLogging.
Run a scheduled check, for example a Lambda function every 15 minutes, that calls GetTrailStatus and alerts when:
IsLogging is falseLatestDeliveryError has a valueLatestDeliveryTime is older than your limitCloudTrail delivers a log file every few minutes, so an hour of silence in an organization trail is not normal. Call the API in the trail's home Region, and pass the full trail ARN if the function runs outside the management account.
Sign in as root in a non-production account on purpose and confirm the message arrives. Repeat after any change to the rules or SCPs. See Do you regularly review your security posture?
If you later connect AWS to a SIEM such as Microsoft Sentinel, rebuild these alerts there and keep the EventBridge rules until each SIEM alert has fired. See Do you send AWS logs to Microsoft Sentinel?