Uncovering Azure's Silent Threats: A Journey into Cloud Vulnerabilities
summary
Investigates Azure Machine Learning and demonstrates concrete failures: storage account keys logged in cleartext in startup and agent environment files, JWTs in URL logged by nginx (captured with tcpdump), and a DSi Mount agent exposes unauthenticated HTTP on 46802 that leaks service logs and sudo commands; uses nmap, IDA, container-escape techniques (based on a Docker-escape tool), and a Meterpreter reverse shell to show privileged container to host escalation. Finds hosts are not shared but jobs can reuse infected VMs; reports led to CVE-2023-23382 and agent hardening.
AI-written summary · about
chapters (8)
- 0:00Azure silent threats
- 1:52Cloud notebook vulnerabilities
- 7:52Clear text credentials on compute
- 13:15JWT tokens in browser logs
- 22:23Spying via Azure ML agents
- 30:53Failed and successful experiments
- 36:13Secure Azure ML deployments
- 39:27Acknowledgements
related talks
52mUncovering Azure's Silent Threats: A Journey Into Cloud Vulnerabilities
57mCloud’s Dirty Little Secret: It Was Misconfigs All Along
50mSecuring Native Cloud Application Perimeter
55mAzure Attacks Unpacked: Real-World Threats & Lessons
48mLateral Movement in the Age of Hybrid Cloud
43mThe Cloud Gray Zone: Vulnerabilities Found in Azure Built-in VM Agents
38mHiding in the Clouds: Abusing Azure DevOps Services to Bypass Microsoft Sentinel Analytic Rules
55mOffensive Azure Security