Definition
A permission system that tracks and limits what an AI agent can do at every step ('hop') as a task is passed between multiple AI agents, ensuring no agent in the chain ends up with more power than the original human user actually granted. Without this, permissions can silently expand as a request passes from one agent to another.
Why it matters
As companies chain multiple AI agents together to complete complex tasks, this closes a structural gap where a low-privilege request could quietly escalate into a high-privilege action several agents downstream.