AI Security in CI/CD LLM Risks, OWASP Top 10 & DevSecOps

Stop prompt injection, poisoned models, leaked keys and invented AI packages before they ship, on one running case
What you'll learn

---

Place each AI risk at the pipeline stage where it enters: editor, pull request, build, test, release or the live app
Decide whether a finding blocks, warns or ships, and say why
Read a scan report, a gate log and a failed eval without panic
Tell scans, evals, red teaming and guardrails apart, and know what each one sees
Understand why a rule written only in a prompt is a request, not a lock, and where server-side checks belong
Tell a useful AI security gate from a noisy one, and who owns each gate on a team
Try reading scan reports, failed eval gates and findings, and triage them in a sensible order
Understand the newer attack paths: AI agents in CI, malware that drives AI tools, and MCP servers as dependencies
Requirements

---

No prior experience needed. No security or machine learning background is assumed.
A rough idea of what a code change or pull request is helps, and Module 1 explains it anyway.
Nothing to install. Every exercise is read and marked on paper or in a text file.
Description

---

This course contains the use of artificial intelligence.
QuickSlice gave away thirty-seven free pizzas because one customer typed a polite sentence to its order assistant. AI security is now a CI/CD problem. The pipeline you already run was built to catch broken code and known library flaws, and it walks past a package an AI coding assistant invented, a model file that runs code when it loads, an API key left in a notebook, a chat message that rewrites the business rules, and a poisoned document the assistant reads. This course gives you the map: where each risk enters the pipeline, which check catches it, and how a team writes a release gate that blocks, warns or ships. One company, one assistant, twelve weeks.
The course follows one example from start to finish. QuickSlice is a pizza delivery chain with fourteen shops, an app, and an AI order assistant called Slice. QuickSlice is a composite case: the company is invented, the incidents are modelled on real public ones, and the numbers are illustrative. You watch its small team launch Slice with zero AI checks and then meet one incident after another. Each incident enters the pipeline through a different door, and each one leaves behind a check that now runs on every change. By the end, the pipeline has a written release gate that says block, warn or ship.
Who this course is for

---

Developers and testers who work on apps with AI features
DevOps and platform engineers who run the CI/CD pipeline
Security engineers who are new to AI apps
Product and engineering managers who approve releases of AI features
Students and career switchers who want a clear picture of AI security in delivery
Not for you if you want a step-by-step setup of one specific scanner, want to train or fine-tune models, or are preparing for a certification exam
Homepage

---