Agentic Payments: How autonomously can AI make payments? | ALLES LEGAL #145

Alles Legal #145

What happens when it is no longer the customer but an AI agent that handles the checkout and payment? In episode #145 of Alles Legal – Fintech-Recht kompakt, we explore which possibilities existing payment law already offers for agentic payments – and where regulatory requirements still limit autonomous transactions. Tune in now!

Podcast: https://paymentandbanking.podigee.io/870-alles-legal-145-agentic-payments-wie-autonom-darf-die-ki-bezahlen/

When the AI agent does not just shop but also pays

The first episode of our Agentic Commerce series looked at how contracts can be formed when an AI agent acts independently on behalf of a user. Now we turn to the next crucial step: payment.

Together with host Dana Wondra from Payment & Banking, Frank Müller, lawyer at Annerton and author at PayTechLaw, discusses the legal framework for agentic payments. At the heart of the episode is a fundamental question: how can a payment be made when the user is no longer actively involved at the point of checkout?

Authorisation and authentication are not the same thing

In a conventional online purchase, the decision to buy, the choice of payment method and the authorisation of the transaction usually happen close together. With agentic payments, these steps may be separated by hours or even longer.

A user might, for example, instruct an AI agent in the morning to book a suitable flight for no more than €300. The airline, final price and recipient of the payment may only be determined later.

From a payments-law perspective, it is important to distinguish between authorisation and authentication. Authorisation means that the customer consents to a payment transaction. Authentication, on the other hand, is used to verify that the person acting is actually entitled to do so.

A broad instruction such as “You may spend up to €1,000 on my trip” therefore does not necessarily resolve all the legal questions surrounding the subsequent payment.

Strong customer authentication meets autonomous payments

This becomes particularly relevant for card payments and the PSD2 requirements on strong customer authentication. For certain online payments, information such as the specific amount and the payee must be linked to the authentication.

That creates a practical challenge for agentic payments: when the user initially gives the AI agent its mandate, these details may not yet be known.

The payment method also matters. Card payments and direct debits are subject to different legal requirements, meaning that the legal assessment of an agentic payment can depend considerably on how the transaction is ultimately executed.

Existing exemptions offer some flexibility

Current payment law is not entirely inflexible. Exemptions from strong customer authentication already exist, including for certain recurring payments, trusted beneficiaries and low-risk transactions.

Some of these mechanisms could also become relevant to agentic payments. However, they do not yet provide a universal solution for fully autonomous transactions.

Why the technical infrastructure matters

At the same time, major payment providers are developing technical infrastructures for agentic payments. These systems can help identify AI agents, document mandates and tokenise payment credentials.

This is highly relevant from a legal perspective. Merchants and payment service providers need to be able to determine whether an agent is genuinely authorised to act for a particular user and whether it is operating within the agreed limits.

Technology cannot solve the legal questions on its own. It can, however, provide the infrastructure required to make mandates, decisions and responsibilities transparent and traceable.

Trust will be key to agentic payments

For Frank Müller, trust is one of the central requirements for agentic payments to become suitable for widespread use. This requires clearly defined mandates, reliable technical infrastructure and unambiguous liability rules.

What happens if an agent exceeds its spending limit? Who bears the risk if it is manipulated? And who is responsible if a technical error triggers an incorrect payment?

Existing payment law already provides a number of starting points. However, Frank Müller sees a particular need for further development around strong customer authentication if AI agents are ultimately to make payments autonomously without requiring the user to approve every individual transaction.

Stay informed and subscribe to the PayTechLaw newsletter. We will let you know as soon as a new episode is available.

About this podcast

Alles Legal – Fintech Recht Kompakt delivers sharp, weekly insights into legal and compliance matters in the world of banking. (in German only)
This podcast is a collaboration between Payment & Banking and PayTechLaw.
Each Wednesday, we unpack the legal developments shaping the financial world – clearly, concisely, and without the legal jargon.
Since 2021, PayTechLaw authors and Annerton attorneys have brought depth and clarity to complex topics.
Whether it’s PSD3, DORA, or FiDA – we provide the legal context you need.
In 20 minutes. No detours.



By continuing, you accept our privacy policy.
You May Also Like
Reporting ICT-Related Incidents under DORA – ESAs Release New Operational Instructions 1
Read More

Reporting ICT-Related Incidents under DORA – ESAs Release New Operational Instructions

On 16 September 2026, the European Supervisory Authorities (EBA, EIOPA and ESMA – together, the “ESAs”) published new “DORA Incident Reporting – Operational Instructions”. The document supplements the existing regulatory framework for reporting major ICT-related incidents with practical guidance on the use of the reporting templates and is intended, in particular, to promote more consistent reporting practices and improve the quality of reported data.
Read More