Suspicious Container Image Export via ctr in Cloud Model Exfiltration

PremiumReviewedSigma · Medium · v1
Product
linux
Category
process_creation
Author
HuntRule
Published
2026-10-20
Updated
2026-10-20

ATT&CK techniques

Discovery → Exfiltration
  1. Recon

  2. Resource Dev

  3. Initial Access

  4. Execution

  5. Persistence

  6. Priv Esc

  7. Defense Evasion

  8. Cred Access

  9. Lateral Movement

  10. Collection

  11. C2

  12. Impact

What it detects

This rule detects use of the containerd ctr client to export a container image to a local tar archive on a Kubernetes node. This behavior matches the ModeLeak attack against Vertex AI where an attacker abused a compromised service agent to pull and export proprietary LLM model images. Exporting images to disk is a precursor to staging and exfiltrating sensitive model or workload data from cloud tenants.

Related detections9 linkedT1613 — drag to rearrange
Suspicious Kubernetes Enumeration via kubectl in Cloud Compromise
Suspicious AWS Transfer Family Server Creation for Data Exfiltration (via cloudtrail)
Suspicious Kubernetes Permission Enumeration via kubectl auth can-i
Suspicious Kubernetes SelfSubjectAccessReview Permission Enumeration
Malicious Kubernetes Reconnaissance Tooling Activity
Suspicious EC2 Snapshot Attribute Modification for External Sharing via CloudTrail
Suspicious Cross-Project Compute Snapshot Creation via GCP Audit
Suspicious Container Environment Reconnaissance in Serverless Build
Suspicious Compute Disk IAM Policy Modification Granting Owner Role via GCP Audit
Suspicious Container Image Export via ctr in Cloud Model Exfiltration
Pivot detection · T1613 · 9 related

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