Agent-to-Agent Delegation

Delegation must narrow authority, never silently broaden it.

Priority: P1
Phase: Phase 2
Primary owner: IAM / Platform
Status: Blueprint

Objective

Multi-agent systems introduce delegated authority chains. Agent B should not gain more authority merely because Agent A asked it to perform a task. The platform needs explicit delegation semantics, depth limits and evidence.

What to build

Represent parent/child agent task relationships and delegated scopes.
Carry original human/service principal through the chain.
Limit delegation depth, duration, tools and data scopes.
Prevent privilege amplification when a child agent has broader standing permissions.
Allow policies based on delegation path and trust level.

Implementation decisions

Use signed delegation tokens with audience and scope.
Compute effective authority as the intersection of delegator rights, delegate rights and policy constraints.
Record reason/task identifier with each delegation.
Expire delegation automatically when the parent task ends.

Definition of done

A child cannot perform an action the delegated task did not authorize.
Security can reconstruct the full chain behind a tool call.
Delegation loops and excessive depth are rejected.

Success metrics

Delegated actions/day
Privilege-amplification blocks
Average delegation depth
Orphaned delegated sessions

CORE RULE

Delegation must narrow authority, never silently broaden it.

This control is specified on page 28 of the blueprint.