When AI pays on behalf of your customer: who actually gives consent?
When an AI agent initiates a payment on behalf of your customer, the most important question is not whether the transaction can technically be processed. The first question is: is the agent actually authorized to do so?
This may sound simple, but it raises a series of new payment-related questions. Who gave the instruction? What authority was granted to the agent? Did that authorization apply to this specific purchase? And what happens if the customer later says, “This isn't what I intended”?
For businesses, agentic commerce is therefore not just a technological challenge. Above all, it becomes a matter of trust, control and proof of authorization surrounding the payment.
A payment is only trustworthy if you know who is acting
In a traditional checkout, the customer completes the most important steps themselves. With agentic commerce, software can take over parts of this process.
Imagine an employee instructs an AI agent to book a hotel. The price must not exceed a certain amount, and the booking must comply with company policy. The agent searches, compares and selects an option.
But as soon as the agent initiates the payment, new questions arise. How do you know the agent is legitimate? How can you verify that the user actually gave permission? And how do you check whether the agent stayed within the scope of that authorization?
This is precisely where payment infrastructure takes on a broader role: not only processing the transaction, but also helping to establish whether the agent, the instruction and the authorization can be trusted.
Not every bot will be unwanted
Merchants are used to blocking bots or subjecting them to additional checks. Agentic commerce makes this distinction more complicated.
Some automated traffic may actually become desirable. An AI agent can act on behalf of a real customer with valid authorization.
This raises a new question: how do you recognize a trusted agent?
Visa is, for example, developing its Trusted Agent Protocol, while Mastercard is focusing on agent registration and verification through Agent Pay. These initiatives demonstrate how the payments industry is preparing for agentic commerce, but there is currently no universal standard. (Mastercard, 2026).
For your fraud prevention strategy, this means that simply distinguishing between humans and bots may no longer be enough. The challenge is shifting towards identifying trustworthy automation: which agent is acting on behalf of a real customer, with valid authorization and within the agreed limits?
eBook
Ready for the next step after omnichannel?
One payment experience, wherever your customer is — online, in-store, and everything in between.
Want to know what unified commerce means in practice for your organisation and where to start? Download this e-book and discover how to future-proof your payment strategy
Consent becomes more specific than simply saying 'I agree'
Consent is also becoming more complex. A customer may authorize one specific purchase. But an agent can also receive broader instructions: only purchase from selected merchants, stay within a set budget or operate exclusively within a particular product category.
The relevant question is therefore not only whether consent was given, but also what exactly was authorized.
For example, Google's Agent Payments Protocol uses different types of mandates to define what an agent is authorized to do on behalf of a user during the purchasing and payment process (Google Cloud, 2025; Visa, 2025).
For you as a merchant, this means you may need more context for each transaction: not just who is paying, but also what instructions the agent received and the limits within which it was authorized to act.
Fraud and disputes also introduce new scenarios
Payments through Agentic AI introduce new risks alongside existing forms of fraud. What happens if someone takes over a legitimate agent account? What if an agent acts outside the agreed limits? Or what if the customer believes the agent misinterpreted their instructions?
Publicly available loss data on these scenarios is still limited. These are relevant scenarios to investigate now, but they are still too new to draw firm conclusions based on real-world data (Visa, 2025).
This makes one question particularly important: what evidence do you need if a transaction is disputed later?
Consider, for example, information about the original instructions, the authorization granted, the agent's identity, authentication and the final order.
Existing payment regulations still apply in Europe
In Europe, agentic commerce does not start from scratch. Existing European payment regulations, including PSD2 and Strong Customer Authentication (SCA), remain part of the regulatory framework governing payments. SCA is based on active authentication by the consumer. When an AI agent independently initiates a payment, this raises new questions about how these requirements should be applied in practice. How existing regulations apply to different forms of agent-initiated payments will therefore need further clarification.
You cannot simply assume that existing checkout and authentication processes will continue to work in exactly the same way. However, claiming the opposite would also be premature: there is still insufficient evidence to conclude that the entire payment infrastructure needs to be redesigned.
Asking the right questions is now more important than immediately building solutions
For businesses, agentic commerce is currently primarily a matter of readiness. You can already start assessing where your organization may be vulnerable or well prepared:
Can you record who gave authorization and what it covered?
Can you make use of existing tokenization and authentication capabilities?
Can you distinguish legitimate agents from fraudulent automation?
Are you capturing sufficient transaction data for refunds and disputes?
Are your APIs, order processes and fraud models prepared for agent-initiated transactions?
In a market where standards, regulations and responsibilities are still evolving, it is important to start identifying the questions you need to ask about your payment infrastructure.
For Buckaroo, this means closely monitoring these developments and exploring what emerging standards around identity, consent, authentication and agent-initiated payments will mean in practice for our merchants and payment infrastructure.
Because when AI starts acting on behalf of your customer, payments will no longer be just a technical step. Above all, they will become a matter of verifiable trust.
Curious which payment solution is right for you?
Compare the options or request a free, no-obligation consultation. We'll help you find the solution that best suits your business, both now and in the future.
