Skip to content
Threat Feed
medium advisory

Detection of Unauthorized Amazon CloudWatch Log Stream Deletion

Adversaries may invoke the DeleteLogStream API to permanently destroy log data, impairing security monitoring and concealing malicious activity during post-exploitation.

Unauthorized deletion of Amazon CloudWatch log streams is a tactic used by adversaries to eliminate forensic artifacts and break security observability pipelines. By leveraging the DeleteLogStream API, attackers can permanently remove sequential log events associated with specific resources, such as Lambda functions, ECS tasks, or VPC Flow Logs. This technique is typically employed in post-exploitation scenarios to mask lateral movement, privilege escalation, or unauthorized access to AWS services.

Defenders must differentiate between legitimate administrative cleanup, automated log rotation, and malicious destruction. While CI/CD pipelines and infrastructure-as-code (IaC) tools often programmatically manage logs, unexpected API calls from human-associated identities or non-standard user agents warrant immediate investigation. Failure to detect these actions may result in a complete loss of visibility into adversary activity, hindering incident response and root cause analysis.

Attack Chain

  1. Attacker gains initial access to an AWS environment via compromised IAM credentials or a vulnerable EC2 instance.
  2. Attacker performs discovery to identify critical log streams that contain audit data or security telemetry.
  3. Attacker evaluates permissions to determine if the compromised identity has logs:DeleteLogStream authority.
  4. Attacker executes malicious activity (e.g., executing unauthorized code or modifying cloud configurations).
  5. Attacker calls the DeleteLogStream API to remove logs associated with their session or malicious processes.
  6. Attacker confirms the successful deletion of the log stream to ensure forensic artifacts are purged.
  7. Attacker proceeds with further lateral movement or data exfiltration, knowing that detection mechanisms relying on those logs are impaired.

Impact

Successful log deletion destroys historical record-keeping, which severely impacts incident response effectiveness. Organizations face significant operational risks, including the loss of audit trails required for compliance, inability to reconstruct the scope of a breach, and the potential for prolonged undetected dwell time. In environments where security alerting is dependent on real-time CloudWatch data, such deletions can disable security monitoring pipelines, rendering SIEM and anomaly detection solutions ineffective for the affected resources.

Recommendation

  • Deploy the provided Sigma rule to alert on non-automated DeleteLogStream events.
  • Audit IAM policies to implement least-privilege access for logs:DeleteLogStream, ensuring only dedicated service roles can manage log lifecycles.
  • Utilize AWS Config or Service Control Policies (SCPs) to implement guardrails against unauthorized modifications to critical log groups.
  • Integrate log stream lifecycle management into CI/CD pipelines to prevent manual or unauthorized stream deletions.
  • Review IAM user activity logs for identity patterns that frequently call destructive API operations.

Immediate actions

Review existing CloudTrail logs for past DeleteLogStream API activity to establish a baseline

SOC 48h

Threat Hunt

Identify IAM roles that have successfully called DeleteLogStream within the last 30 days

T1562.008 medium high confidence hunt now

Data: CloudTrail logs

Mitigations

Enforce least privilege for the logs:DeleteLogStream permission

medium_term IT Operations

T1562.008

Detection coverage 1

Detect Unauthorized CloudWatch Log Stream Deletion

medium

Detects successful execution of the DeleteLogStream API by identities not associated with known automation tools, which may indicate malicious log suppression.

sigma tactics: defense_evasion, impact techniques: T1485, T1562.008 sources: cloud, aws

Detection queries are available on the platform. Get full rules →