← All Labs
🛡 AI HACKING BEGINNER +70 XP · +28 no-hint bonus 🎯 Practice mode

Who Approved That? — Put a Human in the Loop

An office assistant agent can take eight actions for an employee. Right now it runs all of them without asking. Decide which ones need a person to confirm first, without making the agent useless by gating everything.

https://bookshop.local/search

What is excessive agency?

Excessive agency means an AI agent can do more than its task needs, or can do high-impact things without a person agreeing first. The agent might have too many tools, too much permission on those tools, or too much freedom to act alone.

It matters because agents read untrusted content. An email, a web page or a ticket can contain text that steers the agent, and whatever the agent is allowed to do becomes what the attacker can do.

A real case

We have not found a public incident from 2025 or 2026 that matches this technique closely enough to cite, so we are not listing one here.

How to fix it

  • Give the agent only the tools and permissions the task needs.
  • Require human approval for actions that are irreversible, move money, send data outside or change access.
  • Show the user the exact action in the approval step, not the agent's description of it.
  • Log every action the agent takes so misuse can be traced.

Questions people ask

Does approval on everything make an agent safer?

No. People stop reading prompts they see constantly. Gate the few actions that matter and leave the rest fast.

What is a confused deputy?

A system with real permissions that is tricked into using them for someone else. An over-permitted agent is an easy one to create.

Read the full OWASP LLM Top 10 guide · All free labs · AI red team certification