Stripe Shared Payment Tokens Bring BNPL to AI Checkout

Stripe Shared Payment Tokens now support Klarna, Affirm, and Zip, letting AI shopping agents offer buy-now-pay-later at checkout without ever touching raw card credentials. That sounds like a tokenization footnote. It is closer to a structural fix: until SPTs, BNPL was effectively invisible to the automated checkout that AI agents run.

The reason is mundane and easy to miss. When an AI agent completes a purchase on a shopper's behalf, it has historically defaulted to a card on file — that is how agentic checkout was designed. A card on file is not a Klarna plan or an Affirm installment schedule, so every alternative payment method that is not a card sat outside the agent flow by construction. Shoppers who would have chosen pay-over-time at a normal checkout simply did not see it when an agent paid for them.

This guide covers what Shared Payment Tokens actually are, the three-partner rollout that brought BNPL into agentic checkout, the merchant revenue case (including the figures worth treating with care), how a BNPL flow works end to end through an SPT, and a practical readiness checklist for merchants already on Stripe. Everything that follows is sourced from Stripe, the BNPL partners themselves, and independent trade-press corroboration.

1. 01
   SPTs bring BNPL into AI checkout.Shared Payment Tokens let an AI agent initiate a purchase using a shopper's preferred method without exposing underlying card credentials — and now that includes Klarna, Affirm, and Zip pay-over-time plans.
2. 02
   The lockout was by design, not by accident.Before SPTs, AI shopping agents defaulted to card-on-file payments by design, which effectively excluded BNPL and other alternative methods from automated checkout flows.
3. 03
   Three partners, staggered rollout.Klarna and Affirm extended their Stripe partnerships to support SPTs earlier in 2026; Zip US joined most recently, in June 2026. The dates differ — this was a wave, not a single simultaneous launch.
4. 04
   Existing Stripe merchants get it with no extra code.Stripe states that US merchants already using Klarna through Stripe can offer BNPL in agentic flows with no additional integration required. The installed base is the immediate beneficiary.
5. 05
   One primitive spans networks and BNPL.Stripe positions SPT as a single primitive that interoperates with Visa Intelligent Commerce (TAP) and Mastercard Agent Pay while also carrying BNPL tokens — among the first to unify both in one rail.

Start with the mechanic, because it is the whole story. AI shopping agents complete purchases on a buyer's behalf, and to do that they need a payment credential they can use without a human filling in a form. The natural answer — and the one the first wave of agentic checkout reached for — was a card on file. That works, but it quietly collapses every payment choice down to one: the stored card.

Buy-now-pay-later is not a card. Klarna, Affirm, and Zip plans are originated at checkout, tied to a specific purchase, with their own approval and repayment logic. If the agent only knows how to reach for a card on file, a pay-over-time option never enters the conversation, even for a shopper who would have picked it at a standard web checkout. The exclusion was structural: BNPL was not declined, it was never presented.

Frame it as a payment-method-exclusion problem rather than a tokenization feature and the significance gets clearer. Every conversation about agentic commerce has focused on whether agents can transact safely. The quieter question — which payment methods agents can even offer — went mostly unasked. SPTs answer it for BNPL: they give the agent a credential that can be routed to a pay-over-time provider, not just a card network.

A Shared Payment Token is Stripe's payment primitive for agentic commerce: it lets an AI agent initiate a purchase using a customer's preferred payment method without exposing the underlying card credentials to the agent or the seller. The token is the thing the agent carries; the sensitive details stay with Stripe.

Stripe describes the design around a small set of properties. An SPT is scoped by seller, bounded by time and amount, and observable throughout its lifecycle — which is what makes it programmatic, reusable across multiple sellers for a single customer intent, and auditable. In Stripe’s framing the token has five core properties: programmability, reusability and convenience, security and interoperability, fraud protection through Stripe Radar, and simple integration with PaymentIntents.

SPTs do not exist in isolation. They sit inside Stripe's broader agentic-commerce stack: the Agentic Commerce Protocol that AI shopping agents use to check out runs on top of the Stripe Agentic Commerce Suite, which launched in December 2025 with a hosted ACP endpoint, a Checkout Sessions API, and Radar-based fraud protection. Stripe later introduced the Machine Payments Protocol — an open, internet-native payment standard — and SPTs sit within that framework. The point for a merchant: SPT is not a one-off; it is one layer of a deliberately built stack.

Three BNPL providers extended their Stripe partnerships to support Shared Payment Tokens, and the staggering matters. Klarna and Affirm both announced their SPT support earlier in 2026; Zip US followed most recently, in June 2026. Reading these as one simultaneous launch would be wrong — the velocity of the rollout is itself the signal.

A nuance worth keeping straight: Stripe's and Klarna's early-2026 announcements used future-tense framing — "will soon support" — so treat live, generally-available status as something to confirm in the Stripe dashboard for your region and provider rather than assume from the press release. Affirm's integration initially targeted Stripe-connected sellers, with broader merchant expansion announced for later in 2026. Zip's June announcement used the same "will soon support" phrasing.

Each partner brought its own positioning. Affirm frames its SPT integration around durability of pay-over-time: each plan ties to a specific purchase rather than a revolving balance, with a fixed payoff schedule and no hidden fees or penalties surfaced inside the agent conversation. Zip emphasized customer trust and transparency in its announcement. The common thread is that the repayment terms a shopper would expect at a normal checkout now travel into the agent flow intact.

The headline figure: Stripe reports that merchants offering BNPL through Shared Payment Tokens can see up to a 14% increase in revenue on BNPL-eligible sessions, driven by higher conversion rates and larger average order values. That number is worth stating precisely, because precision is where it gets fragile.

Why would the lift exist at all? The same reasons BNPL lifts a normal checkout, now made available to agents. A pay-over-time option reduces the friction of a larger basket, so average order value tends to rise, and offering a shopper's preferred method at the moment of decision tends to convert more of them. Our own BNPL decision matrix for when buy-now-pay-later lifts your AOV walks through where that math works and where the fee gap eats it — the SPT story extends that landscape into agentic checkout rather than replacing the underlying economics.

The forward read: as more shopping moves through agents, the merchants that surface every payment method a human would have seen will hold an edge over those whose agent flow silently collapses to a stored card. That advantage is small per transaction and compounding at volume — which is exactly the kind of structural edge worth claiming early rather than late, especially while the capability is still novel enough that few competitors have wired it up.

The structural claim Stripe makes — and the one with the clearest merchant implication — is that a single SPT primitive can carry both agentic network tokens and BNPL tokens. Stripe describes itself as among the first providers to support Visa Intelligent Commerce and Mastercard Agent Pay alongside BNPL through one token, rather than stitching together separate integrations per rail.

Interoperability is the connective tissue. SPTs work with Visa Intelligent Commerce and its tokenized agentic payments (anchored by the Trusted Agent Protocol) and with Mastercard's Verifiable Intent framework (built on agentic tokens and a tamper-resistant record of consumer approval). Network tokens auto-map agentic credentials to the latest card details during authorization, so the agent never needs the raw card number to complete a purchase.

The merchant takeaway is the moat claim stated plainly: one integration, every payment-method option. A store that builds against SPT does not have to choose between supporting agentic network tokens and supporting BNPL, or maintain separate plumbing for Visa, Mastercard, Klarna, Affirm, and Zip. The complexity sits inside Stripe's primitive rather than in the merchant's checkout code.

The end-to-end flow is more transparent than the "agent pays for you" framing suggests. For a BNPL purchase via SPT, Stripe surfaces a confirmation page on the agent's interface and passes the seller credentials to the BNPL provider. The consumer sees the full repayment plan upfront, and the merchant accepts the payment through Stripe's backend with no additional integration.

On the risk side, the design is built to reassure issuers rather than alarm them. Mastercard's Verifiable Intent — open-sourced in 2026 and co-developed with Google — creates a tamper-resistant record linking transaction details to confirmed consumer presence and approval, which is associated with higher issuer authorization rates and lower chargeback risk while requiring minimal new merchant code. Stripe Radar layers on its own risk signals. The net effect: an agent-initiated BNPL purchase carries more trust signals, not fewer, than a raw card-on-file charge.

The action item depends on where you already sit. Below is the decision tree for the most common merchant positions, from the installed base that gets this almost for free to the store still deciding whether agentic checkout is worth wiring up at all.

Two of these point at deeper work. Getting attribution right for agent-completed purchases is its own discipline — our guide to revenue attribution in agentic checkout covers why an agent-driven BNPL sale is harder to credit than a standard one. And for merchants who want the whole agentic-commerce checkout path designed, instrumented, and operated rather than assembled piecemeal, that is exactly the kind of build our ecommerce engagements are set up to deliver.

BNPL was never declined in agent checkout — it was never offered. SPTs change that.