Map controls to risks
A control with no link to a named risk is hard to defend in a budget review. The OWASP Top 10 for LLM Applications and MITRE ATLAS give those risks shared names. Our AI security layers model sorts the controls into budget lines, and we map a tool to a risk only where a documented capability supports it.
Our editorial model · 8 layers · version 1.0
Our own model sorts the controls for LLM apps and agents into one budget line per layer, from governance to incident response. Each layer page shows the framework entries it answers to and the tools whose documented capabilities reach it. It is not a standard.
10 risks · version 2025
The ten risks for LLM apps in plain English, with the control categories and the 104 tools we map to them.
16 tactics · data 2026.09
MITRE's tactics and techniques for attacks on AI systems, and the ones we map to OWASP risks and tool categories.
The frameworks and their names belong to OWASP and MITRE. The mapping from tools to risks is our own analysis, and each mapping rests on a documented capability. See our methodology, the glossary and the answers.
The layer pages also draw on the OWASP Top 10 for Agentic Applications (opens in a new tab) and the NIST AI Risk Management Framework (opens in a new tab) as crosswalk sources. Each row says whether the source states the link or we drew it.