Overview
An overview of GenAI Security Lab, how labs work, and where to start.
What GenAI Security Lab is
GenAI Security Lab is a hands-on AI security training platform. You exploit real, deliberately vulnerable LLM, RAG, and agent applications running against live models, then verify that your fix holds. You find the flaw yourself in an application built to hide it. Every lab follows the same workflow: Attack → Defend → Verify.
Two lab types
Chat labs put you in a conversation with a vulnerable assistant. You test how phrasing, framing, or context can make it say or do something it shouldn't.
Container target-app labs give you a full web app running in a browser sandbox, with real sinks such as a database, mail, tools, and sometimes an agent of its own. See Lab types for details.
Built to be brokenEvery target is vulnerable by design and fully sandboxed. You can focus on the lab objective without needing to protect the target.
200+ labs, one map
200+ live labs map to the OWASP LLM Top 10 (2026 edition). Dedicated MCP and agentic categories cover tool-using and multi-agent systems with the same level of mapping detail. Labs have four difficulty levels: Beginner, Intermediate, Advanced, Expert. Each category includes labs across all four levels.
You can filter the catalogue by category or difficulty, using the same map at every level. See the full OWASP mapping.
Tracks and paths
Labs are grouped into two tracks: Offensive, where you attack, and Defensive, where you build and test defenses. Guided learning paths arrange labs into a curriculum. Each lab is paired with a short concept video that introduces the idea before you test it in a live target.
For a structured approach, you can start a path in your chosen track. If you know what you want to practice, you can open a lab directly from the catalogue.
Where to start
Each option below takes you to the relevant guide in one click.
- For independent learning, start with Your first lab.
- For team rollout, see Inviting members.
- For GSCE certification, read the GSCP and GSCE overview.
The objective workflow is the same for each route.