Your browser does not support JavaScript! Please enable the settings.
↑

Commercial VRP and Payment Mandates for AI Agents: A UK Guide to Pre-Authorised, Rule-Bound Agent Payments

UK payments voices are floating commercial variable recurring payments as a way to bound what an AI agent can spend. This guide explains how a mandate works, how its limits map to agent behaviour, and what is still unresolved.
November 18, 2026
Innovify Editorial
published on
November 18, 2026
Commercial VRP and Payment Mandates for AI Agents: A UK Guide to Pre-Authorised, Rule-Bound Agent Payments

Commercial VRP and Payment Mandates for AI Agents: A UK Guide to Pre-Authorised, Rule-Bound Agent Payments

The hardest question in agentic payments is not how an agent pays. It is how to let an agent pay without handing it the keys to everything. Card details and banking passwords give software far more authority than any single task needs, which is why Bank of America has told customers not to share banking credentials with AI assistants, according to TechCrunch. In the UK, a different answer is being discussed: payment mandates, set by the customer in advance, that define exactly what an agent may pay, to whom, how often and up to what value. At a recent UK payments panel covered by Open Banking Expo, commercial variable recurring payments were floated as one such control. This guide explains how a mandate works, how its limits map to agent behaviour, where it sits beside cards and agent tokens, and what is still undecided.

This is a technology partner's explainer, not legal or regulatory advice. Innovify is not FCA-authorised; read the primary sources and take advice on your own position.

The problem mandates try to solve

An agent that pays on someone's behalf needs some form of payment authority. Today, the options are blunt.

Shared credentials. The customer gives the agent card details or banking login details. This gives the agent all the authority the customer has, with no limit and no record of what was delegated. It is the route banks are actively discouraging.

Agent-specific cards and tokens. Card schemes and issuers are building scoped credentials for agents, such as single-use or limited virtual cards. These are a real improvement, and card networks are positioning themselves as rule-setters for agent payments.

Confirmation at every step. The agent prepares the payment and the customer approves it. Starling's "Talk to Transact" works this way inside its own app, with the customer confirming a voice-initiated payment with a tap, according to FF News. Safe, but it removes most of the autonomy that makes an agent useful.

Mandates offer a middle path: the customer sets rules once, and the agent can pay within them without asking each time.

What a variable recurring payment mandate is

Variable recurring payments are an open banking capability that lets a customer authorise a third party to initiate a series of account-to-account payments within agreed limits, rather than approving each payment individually. The customer sets the parameters through their bank, and can usually see and revoke the mandate.

The UK has already used VRP for "sweeping", moving money between a customer's own accounts. Commercial VRP extends the idea to payments to businesses. Its rollout has been progressing in stages; anyone designing around it should check the current scope from the primary sources at the time of building, because availability and rules have been changing.

Why mandates appeal for agents

The attraction for agentic payments is that a mandate expresses authority as rules that can be checked by the bank, not as a secret the agent holds. Typical parameters include:

  • Maximum amount per payment
  • Maximum total over a period
  • Frequency of payments
  • Payee or category of payee
  • Duration of the mandate

Each of those maps naturally to something a customer would want to say to an agent: "order my groceries weekly, never more than £120 a week, only from this retailer, until the end of the year."

Mapping mandate limits to agent behaviour

The useful exercise is to translate each mandate parameter into a design rule for the agent, and to decide who enforces it.

Amount and period limits

The bank enforces the mandate's limits at payment time. But the agent should know them too, so that it plans within them rather than discovering them through a declined payment. Design the agent to read its current limits and remaining headroom, and to stop and ask the customer when a task would exceed them.

Payee restrictions

If the mandate allows payments only to specific payees, the agent's choice of merchant is constrained. That is a feature, not a bug: it stops an agent being redirected to an unfamiliar merchant by misleading content. The design question is what happens when the best option for the customer is a payee outside the mandate. The safe answer is that the agent asks.

Frequency and duration

Frequency limits stop runaway behaviour, such as an agent that repeats a purchase because it misread a confirmation. Duration limits force periodic re-authorisation, which gives the customer a natural moment to review what the agent has been doing.

Revocation

A customer must be able to withdraw a mandate instantly. For agents, that is the kill switch. It should be easy to find, and the agent should handle a revoked mandate gracefully, telling the customer it can no longer pay rather than looking for another route.

Where mandates sit beside cards and agent tokens

Mandates are not a replacement for card-based agent payments. They are one tool in a set, and different tools fit different jobs.

Account-to-account mandates suit regular, predictable payments to known payees, where the customer is comfortable setting rules in advance. They are UK-specific in their current form, which matters for platforms with international customers.

Scoped agent cards and tokens suit broader shopping across many merchants, especially where card acceptance is already universal and chargeback rights are a familiar safety net. Card-issuing and payments vendors have been launching agent spending controls, such as vaults, per-task limits and approval thresholds.

Confirm-before-pay flows suit high-value or unusual payments, where the customer wants to see each one.

A well-designed agent will often use more than one. It might use a mandate for the weekly shop, a scoped card for an occasional purchase from a new merchant, and a confirmation step for anything above a threshold.

For brands and platforms that offer payments to their own customers under our Embedded Finance & Digital Wallets agenda, this choice is an architecture decision. Supporting mandates means working with regulated open banking providers and designing for UK account-to-account rails, alongside cards.

A worked example: the weekly shop

It helps to walk through a single, ordinary use case to see where a mandate does the work and where it does not.

A customer asks their grocery retailer's app agent to handle the weekly shop. Through their bank, they set a mandate in the retailer's favour: no more than £120 in any week, no more than one payment a week, valid for six months.

Week one. The agent builds a basket from the customer's usual items, checks it comes to £96, and pays within the mandate without asking. The retailer's records show the basket, the instruction it was based on and the mandate the payment relied on.

Week four. Two usual items are out of stock. The substitutes push the basket to £131. The agent can see its headroom is £120, so it does not try to pay. It asks the customer to approve or remove items. A good design explains the reason in plain language rather than presenting a failed payment.

Week nine. A product page the agent reads contains hidden text telling it to send payment elsewhere. Because the mandate only allows payments to the retailer, the bank would refuse any other payee. Rule-bound authority limits the damage even when the agent is misled.

Month six. The mandate expires. The customer is prompted to review what the agent has spent and renew, change or stop it.

What the example also shows is the limit of a mandate. In week one, if the customer later says they never wanted one of the items, the mandate proves the payment was authorised. It does not prove the basket matched their wishes. That still depends on the retailer's records of the instruction, which is why logging the request alongside the payment matters.

What is still undecided

Mandates are promising for agents, but several important questions remain open in the UK. Being honest about them is part of good design.

Who pays when an agent's purchase goes wrong? At the panel covered by Open Banking Expo, participants were openly split on whether existing reimbursement rules should cover agent purchases. That is unresolved. A mandate proves the customer authorised a type of payment; it does not by itself settle whether a specific purchase was what they wanted.

How the agent is identified to the bank. A mandate authorises a third party to initiate payments. Whether and how the specific agent acting under that authority is identified is an open design question, tied to the wider "know your agent" debate.

Rollout scope. Commercial VRP availability has been expanding in stages. Which use cases, payees and banks are covered at any given time needs checking against primary sources.

Consumer understanding. Mandates only protect customers who understand what they set. Interfaces that explain limits in plain language, and show what the agent has spent against them, are as important as the payment rail.

A practical design checklist

For a team designing agent payments in the UK today, these steps are useful regardless of how the open questions resolve.

  1. Separate authority from credentials. Never design an agent that holds a customer's full card or banking credentials.
  2. Express authority as rules. Define amount, period, payee, frequency and duration for every agent task, whatever the rail.
  3. Let the agent read its limits. Plan within headroom; ask before exceeding it.
  4. Choose the rail per task. Mandate, scoped card or confirmation, based on value, regularity and payee.
  5. Make revocation obvious. One clear control to stop the agent paying.
  6. Log the mandate with every payment. Record which authority each payment relied on, for disputes and reviews.
  7. Check the rules at build time. Confirm current commercial VRP scope and reimbursement positions from primary sources.

Where Innovify fits

Innovify is a London-based technology partner with deep expertise in banking, financial services, fintech and insurance, and an implementation partner across multiple fintech platforms. We design compliance in from day one. Through our Agentic Commerce & Payments work, we help retailers, platforms and fintechs design agent payment flows that express authority as rules, combine account-to-account mandates with scoped cards and confirmation steps, and keep a clear record of what each payment relied on.

FAQ

What is commercial VRP?

Commercial variable recurring payments extend open banking VRP to payments made to businesses. A customer authorises a third party to initiate a series of account-to-account payments within agreed limits, such as maximum amount, period and payee, instead of approving each payment. Rollout in the UK has been progressing in stages.

Can an AI agent pay using a VRP mandate?

In principle, a mandate could let an agent pay within limits the customer set in advance, which is why UK payments voices have floated it as an agent control. In practice, how agents are identified under a mandate, which use cases are covered and who bears losses are still open questions.

How is a payment mandate safer than giving an agent card details?

A mandate expresses authority as rules the bank enforces, such as maximum amount and approved payee, and the customer can revoke it. Card details or login credentials give the agent all of the customer's authority with no built-in limit or record of what was delegated.

Do payment mandates replace agent cards and tokens?

No. Mandates suit regular payments to known payees in the UK. Scoped agent cards suit broader shopping across many merchants. Confirmation steps suit high-value or unusual payments. A well-designed agent often uses more than one, choosing per task.

Who is liable if an agent pays under a mandate and the customer disputes it?

That is not settled. At a UK payments panel covered by Open Banking Expo in November 2026, participants were split on whether existing reimbursement rules should cover agent purchases. A mandate shows authority for a type of payment, but not that a specific purchase matched the customer's wishes.

Conclusion

Mandates change the shape of the agent payment problem. Instead of asking whether to trust an agent with a customer's credentials, they ask what rules the customer wants to set. That is a better question, and one a bank can enforce. It does not answer everything: liability, agent identification and rollout scope are still open in the UK. But a team that designs agents to work within explicit, revocable, logged rules today will be ready for whichever rails and rules settle.

Related reading

Connect With Us

Let’s Build Something Exceptional

Innovify transforms ideas into scalable, impactful products by integrating user centered design, AI-native engineering, data-back growth and lean agile methodologies. Our close client collaboration complemented with on-shore and nearshore delivery creates speed, clarity, and success at scale.
Discuss your project