Unusual AWS Command for a Useredit

A machine learning job detected an AWS API command that, while not inherently suspicious or abnormal, is being made by a user context that does not normally use the command. This can be the result of compromised credentials or keys as someone uses a valid account to persist, move laterally, or exfil data.

Rule type: machine_learning

Rule indices: None

Severity: low

Risk score: 21

Runs every: 15m

Searches indices from: now-2h (Date Math format, see also Additional look-back time)

Maximum alerts per execution: 100

References:

Tags:

  • Elastic
  • Cloud
  • AWS
  • ML

Version: 5

Rule authors:

  • Elastic

Rule license: Elastic License v2

Investigation guideedit

## Config

The AWS Fleet integration, Filebeat module, or similarly structured data is required to be compatible with this rule.

## Triage and analysis

### Investigating an Unusual CloudTrail Event

Detection alerts from this rule indicate an AWS API command or method call that is rare and unusual for the calling IAM user. Here are some possible avenues of investigation:
- Consider the user as identified by the `user.name` field. Is this command part of an expected workflow for the user context? Examine the user identity in the `aws.cloudtrail.user_identity.arn` field and the access key id in the `aws.cloudtrail.user_identity.access_key_id` field which can help identify the precise user context. The user agent details in the `user_agent.original` field may also indicate what kind of a client made the request.
- Consider the source IP address and geolocation for the calling user who issued the command. Do they look normal for the calling user? If the source is an EC2 IP address, is it associated with an EC2 instance in one of your accounts or could it be sourcing from an EC2 instance not under your control? If it is an authorized EC2 instance, is the activity associated with normal behavior for the instance role or roles? Are there any other alerts or signs of suspicious activity involving this instance?
- Consider the time of day. If the user is a human, not a program or script, did the activity take place during a normal time of day?
- Examine the history of the command. If the command, which is visible in the `event.action field`, manifested only very recently, it might be part of a new automation module or script. If it has a consistent cadence - for example, if it appears in small numbers on a weekly or monthly cadence it might be part of a housekeeping or maintenance process.
- Examine the request parameters. These may provide indications as to the source of the program or the nature of the tasks it is performing.