Verify model artifacts, dependencies, licenses, and CI/CD pipeline gates
Most security checklists treat the LLM itself as the risk. This course treats everything around it as the risk.
What you’ll learn
- Identify and scan vulnerable dependencies in an LLM application stack using pip-audit and a structured test case workflow..
- Verify model artifact integrity using SHA-256 checksums and detect unsafe file formats like pickle before they load in production.
- Test LLM behavioral baselines and detect regression when a model is updated, swapped, or fine-tuned without a documented change record..
- Write bug reports and release gate criteria for supply chain findings, including severity classification and remediation evidence requirements..
- Create an LLM-03 QA test matrix with controls, fail signals, owners, evidence, and CI/CD release gates..
Course Content
- Introduction –> 1 lecture • 1min.
- Title Slide –> 4 lectures • 9min.
- Package and Dependency Risks –> 8 lectures • 21min.
- Model Artifact and Provenance Risks –> 14 lectures • 34min.
- Behavioral Regression and Baseline Testing –> 6 lectures • 17min.
- Section 5: Data, Licensing, and Supplier Terms (6 slides) –> 7 lectures • 17min.
- Section 06: AI BOMs, Artifact Integrity, and Release Gates –> 7 lectures • 14min.
- Infrastructure and On-Device Risks –> 5 lectures • 10min.
- Building the LLM-03 QA Test Plan –> 5 lectures • 10min.

Requirements
Most security checklists treat the LLM itself as the risk. This course treats everything around it as the risk.
This course focuses on OWASP LLM-03: Supply Chain. In plain terms, that means testing the components an LLM application depends on before the application ever reaches a user.
An LLM application depends on a software stack you probably did not build. That includes third-party packages, open-source frameworks, model files from public repositories, hosted APIs, supplier terms, and CI/CD pipelines. Any one of those pieces can introduce risk before your prompt logic or application code ever runs.
This course teaches QA engineers how to test the LLM supply chain systematically. You will learn how to scan dependencies for known vulnerabilities, check package and model licenses, verify model artifact integrity with hashes and safer file formats, and build behavioral baselines that catch model drift after a model, adapter, prompt policy, or runtime changes.
You will also learn how to turn those checks into release gates. The goal is not just to find issues. The goal is to capture evidence your team can use for go/no-go decisions, bug reports, risk reviews, and closure criteria.
Every section follows a test-execute-report pattern. You get real tools, runnable demos, and bug report templates you can adapt for your own QA work.
You do not need to be a security specialist. If you can run a Python script, read terminal output, and follow a test plan, you have what you need.
This course works as a standalone. It also pairs naturally with Course 1, Prompt Injection Testing, and Course 2, Sensitive Information Disclosure Testing, in the AI Security Testing series. Together, the first three courses cover the first three entries in the OWASP LLM Top 10 from a QA engineer’s perspective.
By the end of this course, you will know how to look at an LLM feature and ask the release questions that matter. Where did this model come from? Has it changed since the last verified build? Is the artifact identity checked before it loads? Is there a behavioral baseline? If something fails a check, what evidence do we need to route it, report it, and close it?