Log inVerifiable delegated authority for AI agents
The Kordn Delegation Protocol lets organisations delegate bounded authority to AI agents. When an agent acts, its counterparty can verify who authorised it, what it may do, and whether that action is within scope.
An agent's internal permissions stop at the company boundary.
Organisations already use access controls, policies, API credentials, and approval systems to govern software internally. But an external counterparty cannot inspect those controls. When an agent acts across a company boundary, the counterparty needs answers to a different set of questions:
- Who authorised this agent?
- What was it authorised to do?
- What constraints apply?
- Is the authority still valid?
- Does this particular action fall within its scope?
Today, those answers rarely travel with the agent’s action in a form the counterparty can verify.
This keeps the exact structure while making the problem technically cleaner. The final sentence also avoids an overbroad “there is no way” claim.
Identity tells you which agent is acting. Delegation tells you what it may do.
An organisation may know what its own agent is allowed to do. An external organization the agent interacts with do not.
API credentials can authenticate software. Access controls can restrict internal systems. Agent identity can establish which agent is acting. None of these, by themselves, gives a counterparty verifiable evidence that a particular organisation authorised that agent to take this action under these constraints.
That becomes a commercial problem when agents place orders, approve work, negotiate terms, or make commitments across company boundaries.
Who is acting?
What a counterparty can see todayHow a counterparty verifies an agent’s authority
Define the authority
An organisation issues a delegation credential describing what an agent may do and under which constraints.
Bind the action
When the agent uses that authority, it signs the exact consequential action and links it to the credential.
Verify before accepting
The counterparty checks the issuer, credential status, signature, and scope before deciding whether to accept the action.
Using KordnOS, people and systems delegate bounded authority to an AI agent. When the agent acts across the organisational boundary, it signs the exact action and links it to its delegation credential. The counterparty verifies both before deciding whether to accept the action
Designed for real organisational boundaries
- 01
Authority must be explicit and bounded.
- 02
Authority must be verifiable, not merely asserted by the agent.
- 03
Every consequential action must be bound to the authority under which it was taken.
- 04
Delegation must support constraints, expiry, and revocation.
- 05
Each organisation must retain control over its own acceptance policy..
- 06
The protocol must work across vendors, agents, and organisational systems.
Will your AI agents act across company boundaries?
We are working with organisations, agent developers, and infrastructure providers preparing for agents to place orders, approve work, or make commitments across company boundaries.
Technical materials are available to selected deployment partners.
