When an AI agent makes a purchase on your behalf, should it need a completely new payment credential, or should it be able to use the card you've already chosen?
Mastercard has made its preference clear.
During its Q3 2026 earnings call, CEO Michael Miebach backed network tokenisation over solutions that introduce an intermediary, one-time virtual card into an agentic transaction. Mastercard sees these arrangements as a temporary bridge rather than a long-term model.
The reasoning is straightforward. Customers choose their cards for a reason, whether that's rewards, payment preferences or the protections and controls attached to their accounts. Those benefits shouldn't disappear simply because an AI agent is making the purchase.
It's a significant signal about where agentic commerce is heading, and it reinforces a position we've been taking at Thredd.
Agentic commerce doesn't require an entirely new payments system. It requires us to extend the trust infrastructure we already have to support a new kind of payment initiator.
The payment is familiar. The buyer is changing.
The payments industry has been enabling software to initiate transactions for years.
Digital wallets, recurring payments, tokenised credentials and automated authorisation decisions are already part of everyday commerce.
What's different about agentic commerce is the authority we're beginning to delegate to AI.
Today, most agentic purchases still involve the cardholder approving and authenticating the transaction. Over time, agents will increasingly be able to act independently within predefined rules.
Imagine asking an AI agent to find and book a flight for less than £500.
The agent might compare options, select a merchant and initiate the payment using your preferred card. But you would still expect that purchase to respect your spending limits, retain your existing card benefits and follow the instructions you provided.
The challenge isn't simply enabling the transaction. It's establishing that the agent has permission to make it.
For issuers, that changes the fundamental trust question:
Is this the right agent, acting for the right cardholder, within the authority it was given?
Tokenisation protects the credential. But who protects the intent?
Mastercard's emphasis on network tokens is important because tokenisation provides a secure foundation for agent-initiated payments.
By replacing sensitive card details with tokenised credentials, an agent can participate in a payment without needing access to the underlying card number.
It's a natural extension of the approach already used in digital wallets.
But protecting the credential is only one part of building trust.
A valid network token doesn't, on its own, establish whether an agent was authorised to spend £50 or £500, purchase from a particular merchant or complete a transaction without further approval.
That requires additional information about the agent, the cardholder's instructions and the context of the purchase.
It's why emerging agentic payment frameworks are bringing together tokenisation, agent identity, authenticated customer consent and verifiable purchasing intent.
For the issuer, these signals need to support an informed decision at the point of authorisation.
The question is no longer simply whether a credential is valid, but whether the transaction is consistent with the authority the customer delegated.
The issuer becomes the trust anchor
Different customers and card programmes will want different levels of AI autonomy.
One customer might allow an agent to reorder everyday essentials automatically. Another might permit travel bookings within a defined budget. Others may prefer to approve every purchase or prevent agent-initiated transactions entirely.
There is no single risk appetite that works for everyone.
That's why we believe agentic commerce needs more than a simple allow-or-decline decision. It needs a configurable risk envelope, with boundaries around what an agent is permitted to do.
Within those boundaries, agents should be able to act with greater autonomy. When they approach or exceed them, additional authentication or intervention may be appropriate.
Fraud monitoring will also need to evolve.
Machine behaviour doesn't always look like human behaviour. High transaction velocity, unusual purchasing times or multiple rapid transactions may be entirely legitimate for an authorised agent.
Equally, a seemingly ordinary transaction could be a problem if it falls outside the customer's instructions.
Fraud controls therefore need to consider not just whether activity looks unusual, but whether it aligns with the agent's permissions, expected behaviour and transaction context.
This makes issuer-side visibility, controls and real-time decisioning critical to building confidence in agentic commerce.
Building agentic readiness on existing infrastructure
At Thredd, our approach is to extend the capabilities already central to modern card programmes rather than create a separate payment infrastructure for AI agents.
That starts with tokenisation and secure provisioning, allowing approved agents to use protected payment credentials. It extends into authentication, where technologies such as payment passkeys and device binding help establish cardholder identity and consent.
Beyond that, we're developing agent-specific controls and fraud monitoring capabilities to help issuers define spending permissions, recognise agent behaviour and respond when transactions move outside authorised parameters.
Reporting and transaction visibility will also become increasingly important, allowing issuers to identify agent-initiated payments, monitor adoption and refine their controls as the market evolves.
These capabilities build on established foundations.
Thredd has already generated more than 75 million wallet tokens, joined Visa's Agentic Ready programme and, working with Mastercard and PhotonPay, supported live agentic payment milestones powered by Mastercard Agent Pay.
Those developments demonstrate why scheme connectivity, tokenisation and issuer processing are central to agentic commerce readiness.
As new standards and transaction requirements emerge, our role is to help clients translate that ecosystem complexity into practical capabilities, without having to rebuild their payments infrastructure.
From agent-enabled to agent-ready
Mastercard's latest comments are another indication that agentic commerce will evolve through established payment networks, secure credentials and stronger controls over delegated authority.
But network tokenisation alone isn't the end of the story.
The longer-term opportunity is to move from human-assisted purchasing to authenticated agent transactions and, eventually, policy-controlled autonomy.
That transition will depend on bringing together the right information at the point of payment: who the customer is, which agent is acting, what authority was granted and whether the transaction remains within those boundaries.
For issuers, the priority is to prepare for that evolution now, building on infrastructure they already trust.
Because the future of agentic commerce isn't simply about giving AI agents the ability to pay.
It's about giving them permission to act, without giving up control.
Explore how Thredd is preparing issuers for agentic commerce, including the infrastructure, controls and capabilities needed to build trust into agent-initiated payments.
