Suspicious Bedrock AgentCore Runtime Invocation on Wildcard Resources (via aws)

PremiumReviewedSigma · Low · v1
Product
aws
Service
cloudtrail
Author
HuntRule
Published
2026-06-10
Updated
2026-08-28

ATT&CK techniques

Initial Access → Defense Evasion
  1. Recon

  2. Resource Dev

  3. Execution

  4. Cred Access

  5. Discovery

  6. Lateral Movement

  7. Collection

  8. C2

  9. Exfiltration

  10. Impact

What it detects

This rule detects Bedrock AgentCore code interpreter and agent runtime invocations that, in the Agent God Mode scenario, are abused through wildcard IAM permissions to execute code across agent boundaries. Invocation of these runtimes by unexpected principals can indicate exploitation of excessive privileges for arbitrary execution in the AI environment.

Related detections9 linkedT1078.004 — drag to rearrange
Suspicious Azure OpenAI Model Enumeration and Deployment Creation (via azure)
Suspicious Azure Cognitive Services Account Key Enumeration via ListKeys (via azure)
Malicious Business Email Compromise OAuth Application Sign-In via Entra ID Service Principal (via azure)
Suspicious AWS API Call from TruffleHog Credential Scanner User Agent
Suspicious Compute Instance Metadata Modification in GCP
Malicious JavaGhost Security Group Creation (via cloudtrail)
Suspicious S3 Access from Desktop File Transfer Clients by Bling Libra
Suspicious SSH Key Push to EC2 via EC2 Instance Connect
Suspicious EC2 Key Pair Import via aws (via cloudtrail)
Suspicious Bedrock AgentCore Runtime Invocation on Wildcard Resources (via aws)
Pivot detection · T1078.004 · 9 related

Detection content is published as a reviewed draft. Tune thresholds and exclusions against your own telemetry before enabling this rule for alerting.