OWASP Top 10 for LLM apps
Also called OWASP LLM Top 10, excessive agency
A separate OWASP list for apps built on language models. Prompt injection is first. It also covers leaks of private data, poisoned models, trusting model output too readily, and giving a model too much power.
Two entries matter most for agents. Improper Output Handling: model output is untrusted input, so paste it into a query or a page and you're back to injection. Excessive Agency: a model with more tools than it needs.
Say it to your agent
“Go through the OWASP LLM Top 10 for this feature, starting with LLM01, LLM05 and LLM06. Show where model output reaches a query, a page or a shell, and list each tool the model has with the reason it needs it. Remove any without one.”