A chain of delegation is the step-by-step path that shows how authority moves from a human request to an AI agent’s final action, including any policy checks, tool calls, or sub-agent handoffs in between. In an agentic system, one prompt may lead to several internal decisions before anything happens, so the chain explains who or what was authorized at each step.
This matters in cyber security because attackers can exploit weak delegation boundaries to make an agent act outside its intended scope, while defenders need the same chain to prove what happened after the fact. A complete chain supports forensic reconstruction, access control, and accountability: investigators can see which identity issued the request, which component approved it, and which tool executed it. Good defenses use per-agent identities, structured traces, and tamper-evident logs so the delegation trail is hard to forge or hide.