AI Workflows for Cloud Security Testing and Research
· 1:24:41 · Watch on YouTube ↗
Summary
Bo Bulock demonstrates AI-assisted cloud security workflows for tenant enumeration, scanner analysis and access-policy review, and discusses local models, sensitive data and the need for human review.
Key points
- Bulock recommends local models as one way to keep sensitive customer data out of third-party AI services, while noting that local deployment requires suitable hardware.
- A research workflow uses AI agents to triage security articles and produce a focused report; a separate Hermes agent catalogs course material and suggests updates.
- Demonstrated workflows analyze Microsoft tenant enumeration and public cloud storage results, then suggest possible attack paths.
- A multi-agent workflow correlates Prowler findings across cloud services to identify potential exposure and privilege chains.
- Graphrunner exports conditional access policies and OAuth application data for AI-assisted review of potential policy gaps and suspicious applications.
- Bulock warns about hallucinations, incorrect permission logic, stale knowledge and attribution errors, and recommends testing with known data and retaining human review.
Article Details
- Topics
- [06:10] Sensitive security data and hosted AI services
- [11:27] Local models, hardware and cloud-hosted alternatives
- [26:37] Model safety mechanisms and offensive testing
- [28:24] Structured security workflows and research triage
- [36:28] Persistent agents for security research and course updates
- [42:42] Tenant and cloud storage enumeration
- [48:03] Correlating cloud scanner findings into potential attack paths
- [52:04] Conditional access and OAuth application analysis
- [55:54] AI limitations, testing and human review