03

Control plane

Agents, under your mandate.

Cush Core is AI-native. Risk, routing, reconciliation, and first-line operations run as agents inside policy you write. Nothing material happens that cannot be replayed.

Intent and execution

AI reads the intent. The ledger keeps the promise.

Explainable AI helps operators and customers work through what they need, in plain language. In the Cush Core design, money moves only through deterministic smart contracts: the same inputs always produce the same outputs, and every execution is recorded on the ledger. Flexible in how people ask. Exact in what the book records.

QCan a model change what the ledger records?+

No. In the Cush Blockchain design, AI is advisory only. Models explain, summarise and suggest. Postings are made by deterministic smart contracts on the ledger, inside the permissions the institution sets.

Procedure

One instruction, six stations.

A static reading of the control plane. Generic models are widely available. The advantage here is the institution’s data, its policy, and a loop that can pay, flag, or resolve, then show its work.

Not a chatbot on yesterday's ledger. Governed agents that weigh trade offs inside consent and regulatory boundaries, with every decision replayable for the supervisor the same afternoon.

  1. 01

    On-ramp

    The instruction arrives on a named scheme.

  2. 02

    Risk

    Score in flight.

  3. 03

    Routing

    The cheapest viable rail.

  4. 04

    Reconciliation

    Exceptions, closed.

  5. 05

    Operations

    First line, under policy.

  6. 06

    Off-ramp

    Settlement leaves only after the trace exists.

Agents

Policy stays with the bank.

Risk

Score in flight.

Every payment is scored against your policy before it leaves the ledger. Thresholds, typologies, and escalation paths remain yours.

Routing

The cheapest viable rail.

Optional. Where the institution wants it, an agent suggests a rail from Faster Payments, SEPA, Fedwire, SWIFT, CHAPS, TARGET2, FAST, and PIX based on cost, success rate and mandate, inside parameters the institution defines.

Reconciliation

Exceptions, closed.

Postings, acknowledgements, and scheme confirmations are matched automatically. What cannot be matched is queued with a full trace.

Operations

First line, under policy.

Routine support, name enquiry, and case assembly are handled by agents. Anything material is raised to your officers with the replay attached.

Governed agents

Agents answer to the same controls as officers.

In Cush Core, agentic AI is part of the platform, not a chatbot bolted on. Each agent has its own identity, permissions and audit trail, just like a human operator, and sees only the data its permissions allow. Agents act inside the mandate the institution authorises, and every action is recorded.

QAre AI agents governed like staff?+

Yes. Each agent has an identity, a set of permissions and an audit trail. Its capabilities are assigned and reviewed in the agent studio, and it sees only what its permissions allow.

Replay

An illustrative posting.

The bank owns thresholds, typologies, escalations, and the permission an agent has to act. The walk-through below is illustrative. The Cush Blockchain uses BLAKE3.

Replay · illustrative

Every decision, again.

Instruction · 1 / 6

£12,400 received · Faster Payments

Originating institution: corporate in Manchester. Beneficiary: current account at the receiving bank, London. Purpose: supplier settlement.

Hash
1f2582cf1f2582cf1f2582cf
Parent
genesis

The Cush Blockchain uses BLAKE3. This walk-through is illustrative of the replay a supervisor is shown.

Walk an instruction from Faster Payments to the customer’s account, with the agent traces a supervisor will one day request.

Request a briefing