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

Know Your Agent: How Banks and Merchants Can Verify an AI Agent and the Accountable Owner Behind It

Banks and merchants are blocking, detecting and scoping AI agents. This guide explains know your agent: how to verify an agent, bind it to an accountable person or business, and decide what it may do.
November 17, 2026
Innovify Editorial
published on
November 17, 2026
Know Your Agent: How Banks and Merchants Can Verify an AI Agent and the Accountable Owner Behind It

Know Your Agent: How Banks and Merchants Can Verify an AI Agent and the Accountable Owner Behind It

Every bank and merchant now faces the same question when software arrives at the door: is this a customer's sanctioned agent, or something else? For two decades, know your customer has answered "who is this person?". Agentic commerce needs a second answer: "which agent is this, who is accountable for it, and what is it allowed to do?". That second answer is increasingly called know your agent. In November, Deutsche Bank and Ant Digital framed it plainly in a paper on agentic payments in wholesale finance: agents need verifiable identity bound to an accountable owner, according to Finextra's reporting. This guide explains what know your agent means in practice, how it sits beside KYC and consent, and the design choices facing UK banks, merchants and brands.

Why identity has become the first agent question

The first wave of agentic commerce debate asked whether agents could buy things. The second asked who was liable when they bought the wrong thing. The third, now under way, asks how anyone can tell a legitimate agent from an illegitimate one.

The evidence that this matters is in how gatekeepers are behaving. Bank of America told customers not to give banking credentials to AI assistants, and United Airlines barred agents from completing bookings, according to reporting by TechCrunch and PYMNTS. JPMorgan has been reported to block agent card payments, although that report rests on a paywalled summary. In the UK, a vendor-authored article named several large banks as blocking AI agents; those names and figures are unconfirmed and should be treated as indicative until a bank says so itself. At the same time, fraud vendors have launched tools to detect agent traffic.

The common thread is that organisations are willing to let agents read and compare, but not to transact without a route they recognise. Know your agent is the work of building that route.

What know your agent means

Know your agent has three parts. Each answers a separate question, and a credible design needs all three.

1. Agent identity: which software is this?

The first question is technical. Can the bank or merchant reliably identify the agent making the request, as distinct from a browser session, a scraper or a different agent? This is where signed credentials, registered agent identifiers and tokenised access come in. The goal is that an agent presents something verifiable, not merely a claim in a header.

2. Accountable owner: who stands behind it?

The second question is legal and commercial. An agent is not a legal person. Someone, a consumer, a business or the agent's operator, must be accountable for what it does. Deutsche Bank and Ant Digital's framing is useful because it binds identity to that owner: knowing which agent is acting is only valuable if you also know who answers for it.

For a consumer agent, the accountable owner is usually the customer who delegated the task, sometimes alongside the platform that operates the agent. For a business agent in treasury or accounts payable, it is the company and a named person within it.

3. Delegated authority: what is it allowed to do?

The third question is the one that prevents most harm. Even a verified agent with a known owner should only do what it has been authorised to do. That means scopes: which accounts it may read, which payments it may initiate, up to what value, with which merchants, for how long, and with what confirmation.

UK industry voices are already discussing programmable, pre-authorised payment mandates, such as commercial variable recurring payments, as one way to bound an agent's authority, according to coverage of a UK payments panel by Open Banking Expo. Those are panel opinions, not policy, but they show the direction: authority expressed as rules that can be checked, not as a password handed over.

How know your agent differs from KYC

It is tempting to treat know your agent as KYC for software. The analogy helps, but the differences matter for design.

  • Subject. KYC checks a person or business. Know your agent checks a piece of software acting for one.
  • Timing. KYC happens once and is refreshed periodically. Know your agent is checked at every session or action.
  • Main question. KYC asks whether someone is who they claim to be. Know your agent asks whether the agent is genuine, whose it is, and whether this action is within its authority.
  • Failure mode. KYC guards against impersonation and fraud. Know your agent guards against over-reach, redirection and credential misuse.
  • Revocation. For KYC, the end point is account closure. For know your agent, it is instant withdrawal of a scope or token.

The biggest practical difference is frequency. KYC is largely an onboarding event. Know your agent has to be checked continuously, because an agent's authority can change between one request and the next, and because a compromised agent needs to be cut off immediately.

The design choices for banks

For a UK bank, know your agent is part of a wider agent-access policy. The realistic options form a spectrum.

Block. Refuse all agent access to transactional functions. Simple, but it pushes customers towards unsanctioned workarounds such as sharing credentials.

Detect and challenge. Identify agent traffic and add friction, such as step-up authentication, when it tries to transact.

Allow through a sanctioned route. Offer agents a defined way in, typically consented and scoped, with read access first and payments behind confirmation. Open banking's consent model is an obvious foundation for read access in the UK.

Build your own agent. Some banks are building agentic features inside their own apps. Starling has launched "Talk to Transact", which lets customers initiate payments by voice and confirm with a tap, according to FF News. That keeps identity and authority entirely within the bank's own environment.

These are not mutually exclusive. A bank might block unknown agents, allow registered agents to read through open banking, and offer its own in-app agent for payments.

The design choices for merchants and brands

Merchants face the mirror-image question: which agents should be allowed to buy, and on what terms?

The practical starting point is to decide what you need to know before accepting an agent-initiated order. At minimum: which agent or platform sent it, whether the customer behind it is known to you, and what evidence of the customer's instruction exists. That last point links directly to disputes. If an agent-made purchase is later disputed, the merchant will want to show what was delegated and what was asked for.

For brands that run their own wallets or stored-value products under our Embedded Finance & Digital Wallets agenda, the question goes further. If a customer's agent wants to spend from a brand wallet, the brand, working with its regulated partner, needs to decide which agents it trusts, what limits apply and how the customer confirms.

Where the industry is heading

No single know-your-agent standard has won, and it would be a mistake to wait for one. But the direction of travel is visible in what payments companies are shipping.

Control features are becoming products. Fintechs have begun offering customers visible, revocable control over the agents that act for them. Coverage in This Week in Fintech described Mercury's agent controls, a sign-off model at Wealthsimple and browser tokens from Evervault, all aimed at controlled delegation rather than outright blocking. These reports rest on secondary summaries, so treat the details as indicative.

Spending controls are multiplying. Card-issuing, payments and wallet vendors have launched agent-specific spending controls, including vaults, per-task limits and approval thresholds, according to The Paypers. That is a component category, not a standard, but it shows delegated authority being expressed as checkable rules.

Card networks want to set the rules. Commentary in November 2026 described card networks positioning themselves as rule-setters for agent payments. Whatever form that takes, it implies agent identity and authority will be checked inside payment flows, not just at the edge of a bank's website.

Wholesale finance is asking the same questions. Deutsche Bank and Ant Digital's paper applies identity, accountable ownership and bounded authority to treasury and institutional payments. Know your agent is not only a consumer problem.

The practical reading is that the three parts of know your agent, identity, owner and authority, are stable even though the technology implementing them is not. Design for the parts, and keep the implementation replaceable.

What to build first

Know your agent can sound like a standards problem that someone else will solve. Parts of it are; agent identity standards are still emerging and competing. But there is useful work any bank, merchant or brand can do now.

  1. Map where agents already reach you. Which channels see automated traffic today, and which of it is customers' agents rather than bots?
  2. Write an agent-access policy. For each function, decide: block, challenge, allow through a sanctioned route, or serve through your own agent.
  3. Define scopes before standards. Decide what authority an agent could have for each function, expressed as checkable rules: value, frequency, payee, duration.
  4. Name accountable owners. For every agent you operate, and for every category of external agent you accept, decide who answers for its actions.
  5. Keep revocation instant. Whatever credential an agent uses, make sure it can be withdrawn immediately without closing the customer's account.
  6. Log every agent action. Record which agent acted, on whose behalf, under which scope. This becomes your evidence in disputes and reviews.

Where Innovify fits

Innovify is a London-based technology partner with deep expertise in banking, financial services, fintech and insurance, and we design compliance in from day one. For agentic commerce, that means treating identity, accountable ownership and delegated authority as core product architecture rather than a later security add-on. Through our Agentic Commerce & Payments work, we help banks, merchants and brands design agent-access policies, scope models and decision logs, and build them into production systems that a risk team and a regulator can follow.

FAQ

What is know your agent (KYA)?

Know your agent is the practice of verifying which AI agent is making a request, who is accountable for it, and what it is authorised to do. It complements know your customer: KYC establishes who a person is, while KYA establishes which software is acting for them and within what limits.

How is know your agent different from KYC?

KYC checks a person or business, mostly at onboarding. Know your agent checks software acting on their behalf, at every session or action, because authority can change and a compromised agent must be cut off immediately. It also adds delegated authority, the scopes defining what the agent may do.

Why are banks blocking AI agents?

Banks are mainly blocking agents from transactional functions because they cannot yet tell sanctioned agents from unsanctioned ones, or bound what an agent may do. Reports in late 2026 described Bank of America warning customers not to share credentials with assistants. The usual position is to allow reading and comparison but not transactions without a recognised route.

What does an accountable owner mean for an AI agent?

It is the person or business who answers for what the agent does. An agent is not a legal person, so its identity is only useful when bound to someone accountable. For consumer agents this is usually the customer, sometimes with the operating platform; for business agents it is the company and a named individual.

What should a bank or merchant do about agent identity now?

Map where agents already reach you, write an agent-access policy per function, define scopes as checkable rules, name accountable owners, make credentials instantly revocable and log every agent action. These steps are useful whichever agent identity standard eventually prevails.

Conclusion

Know your agent is the identity layer agentic commerce has been missing. It asks three questions: which agent is this, who stands behind it, and what may it do? Banks and merchants that can answer them will be able to move from blocking agents to accepting them on their own terms. Those that cannot will keep facing an uncomfortable choice between shutting agents out and letting them in blind.

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