Executive Summary
In November 2026, researchers at Stony Brook University published a controlled honeypot study demonstrating that cloud credentials embedded in public Docker Hub images are exploited within seconds of publication. Over four months, 300 honeypot images containing AWS, OpenAI, and Anthropic API keys were published across eight container registries. The images were pulled more than 75,000 times, with AWS keys receiving over 140,000 API calls from 5,000 unique IP addresses. The fastest observed credential theft occurred just 7.86 seconds after image publication, with all 100 planted AWS keys abused within 15 minutes. Attackers used static layer analysis rather than running containers, bypassing runtime detection entirely.
This research confirms that automated scanning infrastructure targeting public container registries operates at machine speed, rendering manual secret rotation and post-publication remediation effectively useless. The incident underscores a broader surge in cloud credential harvesting campaigns, as adversaries pivot to supply chain and developer pipeline entry points to gain initial footholds in cloud environments at scale.
Why This Matters Now
Container adoption is accelerating across enterprises, and developer pipelines increasingly publish images to public registries with embedded secrets due to misconfiguration or oversight. With attackers now using automated ContainerBots that extract credentials via static image layer analysis in under ten seconds, traditional detective controls and secret rotation workflows cannot keep pace. Organizations must enforce secrets scanning pre-publication and apply egress controls to detect and block unauthorized API calls originating from compromised cloud identities.
Attack Path Analysis
Automated ContainerBots continuously scan public Docker Hub image layers via static analysis and registry API queries, extracting embedded cloud API keys (AWS, OpenAI, Anthropic) and SSH/database credentials within seconds of publication. Once stolen, the credentials are immediately replayed against cloud provider APIs to enumerate resources, exhaust API balances, and consume cloud services, with the entire cycle from credential exposure to abuse completing in under 15 minutes for all 100 planted AWS keys.
Kill Chain Progression
This analysis maps confirmed threat intelligence to the full cloud kill chain to show where defensive gaps would emerge as an attack progresses.
Initial Compromise
Description
Developers inadvertently embed AWS, OpenAI, and Anthropic API keys along with SSH and database credentials directly into Docker Hub image layers. Automated ContainerBots discover newly published images via Docker Hub's public feed of recently published or updated images, then perform static analysis by unpacking image layers or querying the registry API to extract filesystem contents — all without ever instantiating the container. The fastest observed credential extraction occurred 7.86 seconds after image publication.
MITRE ATT&CK® Techniques
Unsecured Credentials: Credentials In Files
Data from Cloud Storage
Valid Accounts: Cloud Accounts
Cloud Service Discovery
Automated Collection
Resource Hijacking
Cloud Infrastructure Discovery
Transfer Data to Cloud Account
Potential Compliance Exposure
Mapping incident impact across multiple compliance frameworks.
PCI DSS 4.0 – Protection of Stored Credentials and Access Control
Control ID: Requirement 3.4 / 8.6
NYDFS 23 NYCRR 500 – Risk-Based Authentication and Access Privilege Management
Control ID: Section 500.14 / 500.07
DORA (EU 2022/2554) – ICT Security and Access Control / Threat and Vulnerability Management
Control ID: Article 9 / Article 10
CISA Zero Trust Maturity Model 2.0 – Credential Hygiene and Workload Identity Governance
Control ID: Identity Pillar: Credential Management / Application & Workload Pillar
NIS2 Directive (EU 2022/2555) – Supply Chain Security and Access Control Measures
Control ID: Article 21 – Security of Network and Information Systems
ISO/IEC 27001:2022 – Data Leakage Prevention and Authentication Information Management
Control ID: Annex A 8.12 / A 5.17
Sector Implications
Industry-specific impact of the vulnerabilities, including operational, regulatory, and cloud security risks.
Information Technology/IT
Docker Hub credential theft via static image layer analysis exposes IT teams to AWS key hijacking within seconds, demanding Zero Trust Segmentation and Egress Security controls to prevent cloud resource exhaustion.
Computer Software/Engineering
Software engineers embedding cloud API keys in public container images face near-instant credential harvesting by ContainerBots, risking unauthorized AWS, OpenAI, and Anthropic access with exfiltration-weighted kill chain impact.
Financial Services
Stolen AWS and AI API keys enabling resource consumption and lateral cloud movement expose financial services firms to compliance violations under PCI 4.0 and NIST 800-53 SC-7 without Multicloud Visibility controls.
Internet
Public container registry ecosystems face automated mass-scanning threats exploiting unprotected image layers, requiring Egress Security enforcement and Cloud Firewall policies to block unauthorized outbound credential exfiltration at scale.
Sources
- Cloud keys leaked in Docker Hub images are stolen within secondshttps://www.bleepingcomputer.com/news/security/cloud-keys-leaked-in-docker-hub-images-are-stolen-within-seconds/Verified
- ContainerBots Research Project - Stony Brook Universityhttps://containerbots.github.io/Verified
- Thousands of images on Docker Hub leak auth secrets, private keyshttps://www.bleepingcomputer.com/news/security/thousands-of-images-on-docker-hub-leak-auth-secrets-private-keys/Verified
- Over 10,000 Docker Hub images found leaking credentials, auth keyshttps://www.bleepingcomputer.com/news/security/over-10-000-docker-hub-images-found-leaking-credentials-auth-keys/Verified
- Hundreds of leaked AWS keys give full control over corporate accountshttps://www.bleepingcomputer.com/news/security/hundreds-of-leaked-aws-keys-give-full-control-over-corporate-accounts/Verified
- CISA - Securing Software Supply Chains and Container Security Guidancehttps://www.cisa.gov/resources-tools/resources/security-guidance-for-containers-and-kubernetesVerified
- AWS Security Best Practices - Managing Access Keyshttps://docs.aws.amazon.com/IAM/latest/UserGuide/best-practices.htmlVerified
Frequently Asked Questions
Cloud Native Security Fabric Mitigations and ControlsCNSF
Based on the attack progression modeled above, these are the defensive controls that would constrain each stage.
Aviatrix Zero Trust CNSF is directly relevant here because stolen cloud credentials were immediately weaponized to traverse AWS services, enumerate resources, and drain AI platform balances across regions — all behaviors that segmented, identity-aware cloud networking would likely constrain. By limiting which workloads and identities can reach sensitive cloud APIs and lateral service endpoints, CNSF would likely reduce the blast radius even after credential exposure occurs.
Control: Cloud Native Security Fabric (CNSF)
Mitigation: CNSF-enforced workload segmentation would likely reduce the reachability of cloud environments accessed by exposed credentials, constraining how broadly stolen keys could be immediately replayed against cloud provider APIs from arbitrary external sources.
Control: Zero Trust Segmentation
Mitigation: Zero Trust segmentation would likely constrain the scope of IAM privilege escalation paths reachable from workloads, reducing how broadly a single set of stolen credentials could be used to assume roles or enumerate high-privilege policy attachments across the cloud environment.
Control: East-West Traffic Security
Mitigation: East-west enforcement would likely reduce the reachability between cloud services accessed via stolen credentials, constraining an attacker's ability to freely pivot from one AWS service or region to another without traversing a policy-enforced boundary.
Control: Multicloud Visibility & Control
Mitigation: Multicloud visibility and control would likely surface anomalous API call patterns and unexpected traffic origins across cloud provider endpoints, potentially reducing the window in which distributed botnet-style activity could operate undetected across AWS and AI platform APIs.
Control: Egress Security & Policy Enforcement
Mitigation: Egress policy enforcement would likely constrain which destinations and service endpoints cloud workloads are permitted to reach, potentially reducing the volume of unauthorized outbound API calls that stolen credentials could generate against AWS storage and AI platform endpoints.
Even after credential exposure, the residual blast radius across AWS storage and AI platform accounts would likely be meaningfully smaller if segmentation and egress controls had constrained lateral service reachability and outbound API call volumes prior to detection.
Impact at a Glance
Affected Business Functions
- Cloud Infrastructure Access Control (AWS IAM and API Key Management)
- AI and LLM API Budget and Cost Governance (OpenAI and Anthropic API Accounts)
- Container Registry and Software Supply Chain Integrity
- DevOps and Secrets Management Pipelines
Estimated downtime: N/A
Estimated loss: N/A
Exposed credentials include AWS API keys allowing full enumeration and potential consumption of cloud resources, OpenAI and Anthropic API keys used to exhaust paid account balances, and decoy SSH and database credentials. In real-world incidents mirroring this research, organizations face unauthorized cloud resource provisioning, cryptomining charges, AI API quota exhaustion resulting in direct financial loss, unauthorized access to cloud-hosted data stores, and potential lateral movement into internal infrastructure via compromised SSH keys. The median time to first abuse was 51 seconds and all 100 AWS keys were exploited within 15 minutes of publication, indicating near-zero window for remediation once exposed. 5,000 unique IP addresses generated over 140,000 API calls against planted AWS keys, confirming active automated exploitation at scale.
Recommended Actions
Key Takeaways & Next Steps
- • Enforce Egress Security & Policy Enforcement at the cloud perimeter: Deploy CNSF egress controls with FQDN filtering and outbound traffic policy to detect and block unauthorized API calls originating from compromised credentials attempting to reach cloud provider endpoints (AWS, OpenAI, Anthropic) from unexpected sources or workloads — aligning with ZTMM.Data and NIST.800-53.SC-18.
- • Apply Zero Trust Segmentation to workload identities: Implement identity-based microsegmentation policies so that only specific, explicitly authorized workloads and service identities can invoke cloud provider APIs; stolen static credentials used from an unauthorized context will be denied by policy before resource enumeration can occur — aligning with ZTMM.Network, NIST.800-53.AC-6, and PCI.4.0.7.2.4.
- • Enable Multicloud Visibility & Control to detect anomalous API call patterns: Instrument centralized traffic observability across multi-cloud environments to baseline normal API call volumes and flag the mass enumeration pattern (140,000 calls from 5,000 IPs) as anomalous, triggering automated alerting and credential suspension — aligning with ZTMM.CrossCutting.Visibility, NIST.800-53.AU-6, and NIST.CSF.DE.CM-7.
- • Leverage Cloud Firewall (ACF) with AI-based traffic discovery to block credential replay: Deploy cloud-native egress firewall controls with URL/FQDN filtering and AI-driven traffic pattern discovery to identify and block high-velocity, automated API calls to cloud provider endpoints that deviate from established workload behavior baselines — aligning with ZTMM.Network, NIST.800-53.SC-7, and NIST.CSF.DE.CM-8.
- • Use East-West Traffic Security to contain lateral movement after credential theft: Enforce workload-to-workload and service-to-service segmentation policies that prevent stolen credentials from being used to pivot laterally across AWS regions, services, or accounts; restrict inter-service API access to explicitly approved flows — aligning with ZTMM.Network, ZTMM.Data, NIST.800-53.SC-7, and NIST.CSF.DE.CM-1.



