excessive agency
Giving a model-driven system more tools, permissions or autonomy than its task needs, so a fooled or mistaken model can do real damage.
An over-privileged agent turns any injection or mistake into actions such as deletions, payments or data leaks.
In more detail
It ranks third in the OWASP Top 10 for LLM Applications 2026. The cure is a narrower agent enforced outside the model: fewer tools, scoped credentials, a deny-by-default policy on every call, rate limits and human approval for high-impact actions.
Related terms
- least privilege — Giving every program and person only the access they need, and nothing more.
- agent policy engine — Code outside the model that decides, for every proposed tool call, whether to allow it, deny it or ask a person, based on rules written as data.
- human approval — Pausing an agent before a risky action so a person can approve, edit or reject it, then resuming from saved state.
- AI agent — A system in which a model decides its own next steps, calling tools in a loop until a task is done, rather than following a path your code fixed.
Taught in
- Threat-Model An LLM App — AI Engineering: LLM Apps, RAG & Agents
- Prompt Injection — AI Engineering: LLM Apps, RAG & Agents
- Excessive Agency — AI Engineering: LLM Apps, RAG & Agents
- Evaluating RAG And Agents — AI Engineering: LLM Apps, RAG & Agents
References
- OWASP GenAI Security Project: LLM06:2025 Excessive Agency
- OWASP GenAI Security Project: Top 10 for LLM Applications 2026