Concepts for AI Red Teamings
AI security concepts describe what makes AI systems attackable and how to reason about it. Threat modelling maps assets, entry points, attackers and impact for a model-backed system, so testing focuses on realistic risks such as untrusted input reaching a model that holds powerful tools.
Stage: Foundations · Part of the AI Red Teaming roadmap · Last reviewed 19 November 2026
What to learn in Concepts
- Draw the components of an LLM application and mark trust boundaries
- List assets, attackers and likely attack goals for a chatbot
- Explain why models cannot reliably separate instructions from data
- Rank risks by likelihood and impact
- Map findings to a common risk framework such as the OWASP LLM Top 10
Concepts interview questions
What is a trust boundary in an LLM application?
It is a point where data moves from a less trusted source to a more trusted component, such as web content entering a prompt or model output reaching a database. Attacks usually cross these boundaries, so each one needs checks.
How does threat modelling an AI system differ from a normal web app?
The same questions apply, but the model adds a component that follows natural language, so any text it reads can act like input to an interpreter. You must model retrieved content, tool outputs and training data as attack paths too.
Where Concepts fits in the AI Red Teaming roadmap
Concepts belongs to the Foundations stage (intermediate level), which takes about 15 hours. Threat landscape.
Other units in this stage
Back to the full AI Red Teaming roadmap