AI agents force payments industry to verify intent

G+D Netcetera has warned that existing payment authentication models are not sufficient for transactions initiated by AI agents. The industry now needs to verify intent as well as identity.

The shift reflects the rise of agentic commerce, in which software agents can initiate purchases and payments without direct human action. Examples include reordering materials, booking hotel stays, and settling tolls in real time. In these cases, trust moves away from a cardholder's direct approval and towards whether an agent is acting within an authorised mandate.

Current payment security systems were built around what the industry calls a human-in-the-loop model. Strong customer authentication, 3-D Secure, and biometric checks are designed to confirm that a person knows about a transaction and actively approves it.

That model becomes less clear when an AI agent acts on a user's behalf. Identity checks still matter, but they need to be paired with delegated authorisation and verification of the payment's context and purpose.

Changing fraud signals

According to G+D Netcetera, many signals used in conventional fraud detection lose value in agent-led transactions. Device data, behaviour patterns, and location checks are less useful when the payer is not personally carrying out the purchase.

In their place, a different set of questions emerges. Payment providers and issuers need to know whether the agent's identity is known, whether the mandate remains valid, and whether the purchase falls within the authority granted to that agent.

The broader payments market has begun to address these issues, although the framework remains incomplete. E-commerce platforms are testing autonomous customer journeys, while card networks have introduced specifications for agent-driven payments, including Visa Trusted Agent Protocol and Mastercard Agent Pay.

Visa has also reported live agent transactions with European merchants, supported by more than 30 issuing banks. That suggests practical deployment has begun, even as standards, liability rules, and supervision remain unsettled.

Regulatory gap

A central issue is regulation. Under PSD2, strong customer authentication assumes that the person making the payment initiates the transaction directly. That creates uncertainty for payments made by an agent acting within a previously approved mandate.

The result is a grey area for banks, merchants, and payment service providers. If an AI agent buys the wrong item while still operating within the formal boundaries of a mandate, the dispute may not qualify as fraud in the traditional sense.

That distinction matters because existing chargeback rules do not clearly allocate losses in such cases. A transaction can be valid from a technical and authentication perspective while still being challenged over whether the agent properly reflected the user's intended scope.

Work on standards is moving forward. The Agent Payments Protocol, or AP2, has been handed to the FIDO Alliance, where a working group chaired by Mastercard and Visa is developing specifications for agent-initiated commerce.

G+D Netcetera said this approach would make the chain of proof cryptographically verifiable. It also points to a broader industry effort involving payment security specialists, infrastructure providers, identity experts, issuers, and payment service providers, rather than only AI companies.

Role for banks

For financial institutions, the rise of agentic commerce could create a bigger role in managing mandates and trust. Banks and issuers already verify identities, manage permissions, and secure payments, putting them in a natural position to act as credential providers for AI-driven purchasing.

That opportunity also brings operational and legal risk. Institutions may need auditable processes that record who granted authority, what limits were attached to it, and whether an agent remained within those terms at the point of transaction.

G+D Netcetera argued that the transition does not require a wholesale replacement of existing payments infrastructure. Tokenisation, EMV 3-D Secure, and FIDO-based authentication can be extended with delegated authorisation, continuous real-time risk assessment, and context-based decision logic.

What matters, in the company's view, is combining those tools into processes that are consistent and traceable. That would allow issuers and merchants to assess not only whether a transaction was authenticated, but whether it was appropriate to the authorised task.

Ivan Vukelikj, Secure Digital Payments Expert at G+D Netcetera, described the scale of the shift: "AI agents will fundamentally change payments. With every delegated purchase decision, the question of how trust can be established between people, AI agents, and payment infrastructures grows in importance. Those who lay the groundwork today for authorized, traceable, and context-based transactions are laying the foundation for the next stage of digital payments."