Agentic Payments. Who Holds the Off Switch?
Ai Agents are starting to move Money. The question that matters is not how clever they are, but who answers for them.
Your Ai assistant can already find you a flight, compare prices and fill your basket. The next step is the one that matters: it pays.
The moment an agent moves money, three questions stop being theoretical. Who was it acting for? What was it allowed to do? And when something goes wrong, who answers for it?
The Right Questions are now being Asked
In a speech on 19 October 2026, the FCA set out four elements for a framework for agentic finance: agent identity, delegated authority, accountability and execution. It is a clear and practical frame, and it deserves practical answers.
Earlier this week I contributed a short practitioner paper to the FCA AI Lab’s work in this area. It draws on something unusual: building a commerce platform run by one founder and 21 specialised Ai agents. If any business tests the question of how much can be handed to software, and what must stay with people, it is ours.
These are the ideas behind that paper. They start from one principle:
An agent should never be the thing that decides whether money moves.
1. Identity is more than a Name
An agent’s name tells you almost nothing. Before anyone accepts a payment instruction from an agent, they should be able to answer four questions:
- Who is it acting for? A real person or business, not just an app.
- Who runs it? The firm that stands behind it.
- What is it running on? The model and the version.
- What is it allowed to do right now? Its current mandate and limits.
The third question matters more than it looks. A system that behaved well yesterday can behave differently when the model beneath it changes. If the version is invisible, nobody can judge that risk.
2. Authority should come with an Off Switch
The person should hold the off switch, and it should work instantly. Everything else follows from that.
At KAELUM, people choose how much each capability can do: watch and report, recommend and wait, or act within limits they set. The default is the lowest level. Any step up should need a fresh, specific yes, not a setting buried in a menu. Revoking authority should take effect before the next instruction, not at the next settlement cycle.
We put it in one line: K.A.T.E. governs the Rails. You govern the Spend.
3. Accountability Belongs to People
We describe K.A.T.E. as KAELUM’s Ai Chief Operating Officer. It is a useful way to explain what it does. But a model cannot be accountable, and no job title changes that. Every function our agents perform maps to a named person, and in our case that person is me.
Accountability also needs evidence. Sensitive actions pass a human approval gate before they happen, and governed decisions are written to a tamper-evident record. When something is reviewed later, what is reviewed is what actually happened.
4. When in doubt, Money waits
Model availability should never decide whether money moves. In KAELUM, clear rules decide; the Ai explains, monitors and assists. If any part of a check is unavailable, the payment does not slip through unscreened. It waits for a person.
This is also how we think about relying on a single Ai model. We chose one model family deliberately, because one consistent intelligence gives consistent behaviour and a consistent audit trail. We manage the concentration risk by design: no payment depends on the model being available.
5. Make good Governance possible for Small Firms
Continuous assurance is the right goal, but it has to be achievable by a firm without a risk department. If only banks can afford to prove their agents are safe, agentic finance will be built by incumbents alone.
Our suggestion is a short, standard evidence set that any firm using agents in payments could keep and show on request:
- A register of agents: role, operator, model and version, and autonomy level.
- A record of the authority each agent held, for whom, and when.
- A map of each agentic function to a named, accountable person.
- A plan for what continues, what pauses and who acts when something fails.
- A log of near misses, reviewed on a fixed cycle.
For a small firm these are documents and tables, not new systems. They give a supervisor the same view of a two-person start-up as of a large institution.
Governance First, then Scale
KAELUM is pre-launch, and that is exactly why we built the governance first. It is far easier to design accountability in than to bolt it on after the agents are already moving money.
The firms that earn trust in agentic finance will not be the ones with the cleverest agents. They will be the ones who can show, for every payment, who it was for, what was allowed and who answers for it.
Crypto is powered by Blockchain. KAELUM is powered by Ai.
If you are building with agents, which of these four questions is hardest for you to answer today?
Reference: Building the future of finance in the age of AI, FCA speech, 19 October 2026
KAELUM Technologies Ltd is registered in England and Wales (Company No. 16681154). KAE is a closed-loop commerce currency fixed at £0.20 per unit. It is not a crypto asset, security, deposit or investment instrument, does not appreciate in value, pays no interest or return, and cannot be traded. Only KAE earned through commerce by merchants and creators may be redeemed for fiat, subject to fees and limits. KAELUM is pre-launch and is working towards the appropriate regulatory permissions before it takes any customer funds. Third-party and public-body information is drawn from published sources as of October 2026 and may change, and its inclusion does not imply endorsement of KAELUM by any named organisation, including the FCA. This article is for informational purposes only and does not constitute financial, legal or investment advice, or an invitation or inducement to invest.