Agent Authority Chain

Authority should travel with the agent

Verifiable Authority Chains for AI Agents

Agent Authority Chain makes human intent, delegated authority, and narrowing constraints explicit — so an AI agent can prove not only who it is, but what it is allowed to do.

See how it works View project status

Current boundary Research and validated local demo. Not a production authorization service.

AUTHORITY TRACELOCAL DEMO
01
PRINCIPALHuman

Intent: book travel

SIGNED
delegate≤ €5,000
02
DELEGATEPersonal agent

Capability: travel.*

VERIFIED
narrow≤ €2,000
03
DELEGATETravel agent

Capability: travel.book.hotel

VERIFIED
execute€1,800
04
TRANSACTIONWithin authority

Intent and constraints preserved

VALID

The missing layer

Identity tells us who.
Authority tells us whether.

AI agents are beginning to act across services, tools, and other agents. Identity alone cannot answer the questions that matter at execution time.

  • 01Who authorized this action?
  • 02What exact capability was delegated?
  • 03Were constraints narrowed or silently widened?
  • 04Does the final transaction still match human intent?

How it works

A chain that can only get narrower.

Each delegation carries the authority it received, the limits it must preserve, and the evidence needed for deterministic verification.

01

Bind intent

Start from an explicit human purpose instead of inferring permission from identity.

02

Delegate authority

Pass a precise capability to an agent, with bounded scope, depth, time, and constraints.

03

Narrow, never widen

Every child delegation must remain inside the authority granted by its parent.

04

Verify before action

Reject broken links, widened constraints, expired authority, or intent mismatch before execution.

Where it matters

Built for multi-agent action.

COMMERCE

Agentic transactions

Prove that a purchase, booking, or procurement action stayed inside an approved budget and purpose.

TOOLS

Delegated tool use

Constrain what an agent may do across APIs, MCP tools, and services without granting blanket access.

ORCHESTRATION

Agent-to-agent delegation

Keep authority traceable as a primary agent assigns narrower tasks to specialist agents.

Current status · v0.4.0

The core verification model is running.

A local demo validates plain and signed authority chains, narrowing constraints, request scope, expiration, and intent preservation. The next work is turning the verified model into a carefully bounded service.

Reference checks
19 / 19 passing
Signed envelope
Ed25519 demo
Runtime
Validated locally
Public API
Not exposed

A new primitive for accountable agents

Identity is the start.
Authority is the proof.

Agent Authority Chain is being developed as experimental, self-hostable infrastructure for verifiable agent workflows.