For most of the last decade, the card networks have answered the same question the same way: an interchange fee is an interchange fee, regardless of who pressed the button. That position is, quietly, beginning to change.
What the memos say
According to two internal pricing memos seen by Fintechly — one circulated inside Visa in February, the other dated six weeks later from Mastercard — both schemes are now preparing a separate, higher take-rate for transactions initiated by autonomous software agents. The proposed premiums vary by country and by issuer relationship, but the direction is consistent: somewhere between 9 and 22 basis points above the equivalent card-not-present rate.
A small group of large acquirers and platforms — people familiar with the discussions named Stripe, Adyen, Checkout.com and one of the largest US travel platforms — have been briefed on the framework over the last quarter. None of the companies named would comment on the record. Visa and Mastercard both declined to discuss specifics.
The schemes' risk argument
The schemes' argument, set out in both memos in slightly different language, is one of risk. Agent traffic, they say, behaves nothing like a card-on-file recurring charge: the device is unattended, the authentication signal is weaker, the dispute pattern is unproven. Higher fraud, higher chargebacks, higher cost to acquire each new permission token. They want to be paid for it.
"The networks have spent eighteen months telling regulators that AI-agent commerce is the next category. They cannot now act surprised that they are pricing it like one."
The acquirers' response
The acquirers' response, in private, has been less philosophical. One senior commercial leader described the proposal as "the first time in fifteen years that the schemes have invented a fee category from scratch and tried to anchor the price before anyone has the data to argue about it". The acquirer side of the table will, predictably, lobby for a much narrower definition of what counts as agent-initiated — and a smaller premium for it.
What we know — and what we don't
The memos are clear on three things and silent on most of the rest. They are clear that the premium is intended to be on top of, not in place of, existing scheme and interchange components. They are clear that the launch window is the second half of 2026, in selected corridors first — the United States, the United Kingdom, Singapore. And they are clear that the schemes intend to publish a public methodology before any production pricing goes live.
What the memos do not specify is how either scheme intends to actually identify an agent-initiated transaction at the network. There is no obvious technical signal in the existing message format. Several people in the discussions described an evolving negotiation around a new merchant-supplied flag, audited periodically. If that flag becomes the basis of pricing, the integrity of the audit will become one of the most contested questions in the industry.
What comes next
The conversation is, for now, deliberately quiet. None of this has been put in writing to any acquirer board. It will be, in the next quarter, when the first pricing pilots begin. Regulators in the three launch jurisdictions have been informally briefed; none have yet asked a written question. They will.
Fintechly will continue to report on the methodology, the first pilot corridor, and the regulatory response. Readers with direct knowledge are invited to contact the newsroom.
