How payment processors are rebuilding identity, authorization, checkout, fraud and merchant integration for autonomous agents

The answer in one table

Source note: 2025 scale metrics use each company’s own reporting terminology and are not perfectly comparable. Stripe reports volume generated by businesses on Stripe; Adyen reports processed volume in euros; Checkout.com reports total payment volume. [1][7][11]

Figure 1. Agentic checkout changes the trust problem from “is this card valid?” to “is this agent, mandate and transaction valid together?”

AI checkout creates a new payments problem: authenticated does not mean authorized

A normal ecommerce payment has an unusually clear moment of intent: the shopper adds an item to a cart, sees the final price and presses a button. The payments stack then asks familiar questions. Is the credential valid? Does the issuer approve? Does the transaction look fraudulent? In agentic commerce, those questions remain, but they are no longer sufficient.

An AI agent can be legitimate, the underlying card can be legitimate, and the customer can be legitimate while the purchase is still wrong. The agent may have selected the wrong merchant, exceeded a budget, used an expired instruction, accepted a delivery window the customer did not intend, stacked a promotion in a prohibited way or simply misunderstood a natural-language request. Payments infrastructure therefore has to prove three things separately: who the user is, which agent acted, and whether the agent stayed inside the user’s mandate.

That is why the strategic contest between Stripe, Adyen and Checkout.com is not mainly about adding an “AI checkout” button. It is about who can make delegated commerce auditable enough that merchants, issuers and consumers will trust it at scale.

Stripe: the deepest agent-native stack today

Stripe has taken the most vertically integrated approach. Its Agentic Commerce Suite spans merchant discovery, catalogs, delegated checkout, payment credentials, agent wallets, machine-to-machine payments and fraud controls. Rather than bolt AI onto the final authorization step, Stripe is trying to become the operating layer between the AI interface and the merchant.

The most important primitive is the Shared Payment Token, or SPT. Stripe designed SPTs so an agent can initiate a payment without receiving the underlying card credential. More importantly, the token can be scoped to a particular merchant, amount and time window, revoked, and monitored through its lifecycle. That is a direct answer to the “permission fraud” problem: the payment credential itself can carry constraints around what the agent is allowed to do. [2]

Stripe is also broadening the authorization layer beyond its own token. It says SPTs can carry network-led agentic capabilities from Visa and Mastercard as well as BNPL tokens from providers such as Affirm and Klarna. This matters because merchant adoption is likely to be slowed if each agent, wallet, card network and financing method requires a different integration. Stripe’s strategy is to make one credential abstraction sit above them. [5]

The checkout layer is similarly ambitious. Merchants can publish product catalogs and business profiles to Stripe, expose agent-ready checkout, keep their existing pricing, tax, shipping, refunds and disputes processes, and remain the merchant of record. For AI platforms, Stripe offers ways to embed checkout without becoming the merchant or handling PCI-sensitive card data. [1][6]

Stripe also has a distribution advantage that is hard to replicate quickly. Businesses running on Stripe generated $1.9 trillion in volume in 2025, and Stripe says it directly or indirectly powers more than five million businesses, including most of the largest U.S. technology companies. It is already deeply integrated with the developer ecosystem that is building AI agents. [7]

Where Stripe still has risk

Vertical integration is useful until it becomes lock-in. A merchant that relies on Stripe for catalog distribution, checkout, credential delegation, fraud scoring, wallets and payments gets a coherent stack, but also increases dependence on one provider’s protocols and risk models.

There is also a subtle liability question. SPTs can prove that an agent possessed a scoped credential, but a future dispute may hinge on whether the original human instruction was itself clear and whether the merchant’s final cart still matched that instruction. Token design reduces ambiguity; it does not eliminate semantic ambiguity in natural-language mandates.

Figure 2. Publicly announced agentic-commerce architectures as of October 7, 2026.

Adyen: the “universal translator” for enterprise commerce

Adyen’s strategy starts from a different assumption: agentic commerce may remain fragmented for years. Instead of betting that one AI platform or one protocol becomes dominant, Adyen Agentic is designed to translate between them. Its three public modules are Agentic Feed, Agentic Cart and Agentic Payments. The pitch to enterprise merchants is simple: integrate once, keep the merchant relationship and let Adyen absorb the protocol churn. [8]

That is especially attractive for large retailers whose commerce stack includes ERP systems, tax, inventory, fulfillment, loyalty, stores and existing payment infrastructure. Adyen’s Cart layer is designed to turn AI-generated purchase intent into a live merchant cart with current pricing, tax and shipping, rather than allow the agent to invent or cache the final order state. That helps reduce a key source of agent error: stale or incomplete commercial data.

Adyen may also have the strongest identity narrative of the three. Its Dynamic Identification layer draws on trillions of online and in-person interactions, and the company explicitly argues that agentic commerce requires identity, delegation and intent to be modeled separately. Its 2026 fraud report says merchants must distinguish legitimate agents from malicious ones, verify the agent’s authority to act, and test whether the action remained within the customer’s intended scope. [9]

The underlying payments scale is substantial. Adyen processed €1.394 trillion in 2025 and €803.8 billion in the first half of 2026. Its single-platform model connects acquiring, risk, authentication and payment data across online and physical commerce, which may give it a particular advantage for omnichannel retailers whose customers use AI to discover online but fulfill, return or service purchases in stores. [11][12]

Adyen’s trade-off: enterprise strength can mean slower distribution

Adyen’s architecture is compelling for enterprises, but it does not currently have the same direct AI-developer distribution or consumer payment surface as Stripe’s Link ecosystem. Its edge depends on merchants valuing control and interoperability more than the convenience of a vertically integrated agent stack.

That makes Adyen especially well positioned if agentic commerce develops the way enterprise payments usually develop: fragmented, regulated, region-specific and full of legacy systems that must be connected rather than replaced.

Checkout.com: performance infrastructure first, agent surface second

Checkout.com is building for the same future, but its public strategy is more payments-performance-centric. It supports the Agentic Commerce Protocol, emphasizes merchant-controlled checkout, and is explicit that agent identity, scope and permissions should become new trust signals. Its own agentic-commerce guidance says liability still follows existing payment models: the merchant remains merchant of record and is still responsible for fraud, chargebacks and disputes. [13][14]

This approach is less flashy but could be durable. Checkout.com already has deep infrastructure around network tokenization, 3DS, risk scoring, fraud rules, Intelligent Acceptance and disputes. For an agentic transaction, those tools can be fed a new class of data: which agent initiated the payment, what scope it had, which limits applied and whether the final transaction was consistent with those permissions.

Checkout.com is also unusually explicit about the post-transaction problem. Its agentic-chargeback material argues that traditional chargeback rules were built around a visible “pay” click and that agent-driven commerce needs richer, verifiable evidence of delegation. That focus matters because merchant adoption will slow quickly if agent traffic produces higher fraud or dispute rates that cannot be defended. [15]

The company is smaller than Stripe and Adyen by reported 2025 volume, but it is growing quickly. Checkout.com processed more than $300 billion in 2025, up 64%, and now expects roughly $480 billion for full-year 2026. It also says it works with more than 1,000 enterprise merchants and processed nine billion transactions in 2025. [16][17]

Checkout.com’s trade-off: excellent rails, less complete public agent stack

The current weakness is breadth. Stripe has already productized a much larger agent toolkit, while Adyen has packaged a clear Feed–Cart–Payments architecture. Checkout.com’s strengths are visible in the underlying components—authorization, tokenization, fraud, chargebacks and enterprise acquiring—but its public agentic suite is less unified today.

That does not make it strategically weak. If merchants ultimately insist that AI platforms hand control back at checkout, then the competitive advantage could shift away from owning the agent interface and toward optimizing the ordinary payments stack underneath it. Checkout.com is positioned for that version of the market.

Figure 3. Adoption interest is rising faster than trust. Checkout.com says 42% of merchants are already testing agentic commerce and 76% believe PSPs will be decisive to scalability.

Five layers where the real competition will be decided

1. Agent identity: can the processor tell a sanctioned agent from a hostile bot?

Traditional fraud engines often treat automation itself as suspicious. Agentic commerce reverses that assumption: some bots are authorized customers. Adyen is furthest in publicly articulating this identity problem through Dynamic Identification and its fraud research. Stripe approaches it through SPT lifecycle data and Radar. Checkout.com is building agent identity and behavior into its risk model, but today its public tooling remains closer to conventional risk scoring.

2. Delegated authorization: can permission be expressed in machine-readable form?

This is Stripe’s strongest layer. SPTs are explicitly programmable around seller, time and amount, while Link and Issuing give agents controlled ways to pay. Adyen is emphasizing merchant-owned mandates and collaboration with network token standards. Checkout.com’s framework recognizes agent scope and permissions, but has disclosed fewer proprietary delegation primitives.

3. Checkout integrity: can the final order stay synchronized with merchant reality?

Adyen’s Agentic Cart directly targets the problem of stale price, inventory, tax and shipping data. Stripe’s Delegated Checkout does the same from a more developer-native angle. Checkout.com increasingly argues for merchant-controlled checkout, especially as OpenAI has shifted away from a universal in-chat Instant Checkout model. In practice, this layer may matter more than the payment credential: an accurately authorized purchase can still be wrong if the cart is wrong.

4. Fraud and authorization: can AI traffic convert without becoming a fraud loophole?

All three have credible systems. Stripe has Radar and a huge internet-commerce dataset. Adyen combines acquiring, identity and risk on a single stack and says nearly all top customers use Uplift; its H1 2026 results said Uplift and Dynamic Identification improved customer conversion by 0.9 percentage points on average. Checkout.com has risk scores, Fraud Detection Pro, network tokens and an AI optimization system focused on authorization performance. No public evidence yet demonstrates a decisive agent-specific fraud-performance winner.

5. Disputes and evidence: can a merchant prove what the customer meant?

This will be underestimated until autonomous purchases become common. A future chargeback may ask not only whether the cardholder authenticated, but whether the agent was permitted to buy this item from this merchant at this price at this time. Stripe’s scoped token lifecycle creates useful evidence. Adyen wants merchants to own the mandate and proof. Checkout.com is especially explicit about building richer dispute evidence around delegation. The company that makes this evidence portable across issuers and networks could create a powerful moat.

What “best positioned” means depends on the merchant

A scale thought experiment: even modest agent penetration becomes huge

Adyen’s 2026 fraud research cites industry expectations that agents could influence roughly 5%–20% of payment volume within five years. Applying that range mechanically to each company’s reported 2025 scale shows why the strategic stakes are large. This is not a forecast and the three volume measures are not perfectly comparable.

The implication is not that processors will earn a new fee on all of this volume. It is that agentic commerce can change who controls the checkout, who owns the consumer credential, which fraud signals survive the handoff, and which processor becomes the default abstraction layer across AI platforms.

The biggest strategic risk is not fraud. It is losing the merchant relationship.

The obvious risk in AI checkout is unauthorized purchasing. The less obvious strategic risk is disintermediation. If an AI platform owns discovery, product comparison and the consumer interface, the merchant may be reduced to a fulfillment endpoint. Payment processors therefore have a second job: preserve the merchant’s identity, economics and post-purchase relationship even when the user never visits the merchant’s website.

All three companies increasingly emphasize this point. Stripe says the merchant remains the merchant of record. Adyen makes merchant control a central design principle. Checkout.com has welcomed the shift toward merchant-controlled checkout. This convergence suggests the industry has already learned an important lesson: AI shopping is more likely to scale if it behaves as a new commerce channel, not as a replacement for the merchant’s legal and operational stack.

That also explains why the payment processor may become more valuable in the AI era rather than less. The processor is one of the few actors that can sit simultaneously between the merchant, the agent, the card network, the issuer, the credential, the fraud model and the dispute process. In agentic commerce, orchestration is the product.

Verdict: Stripe leads the product race; Adyen may lead the architecture race

If “best positioned” means which company has the most complete agentic-commerce product stack available to developers and merchants today, Stripe has the clearest lead. It has built proprietary primitives at nearly every layer and can distribute them through an enormous base of internet businesses and AI companies.

If “best positioned” means which company has the most defensible architecture for a fragmented enterprise market, Adyen may be equally compelling. Its universal-translator thesis assumes no single AI platform, token standard or protocol will dominate. That is a conservative assumption—and historically a reasonable one in global payments.

Checkout.com is not out of the race. Its edge is that agentic commerce will still have to survive the same economic realities as ordinary ecommerce: approval rates, false declines, fraud losses, chargebacks, reconciliation and merchant control. If the market standardizes around external protocols and merchant-owned checkout, performance infrastructure could matter more than proprietary agent primitives.

The most likely industry structure is therefore layered rather than winner-take-all. Stripe may dominate AI-native commerce surfaces; Adyen may become the translation layer for large enterprises; Checkout.com may win merchants where payment performance and dispute economics dominate the purchasing decision. The AI checkout will be less about who has the smartest agent and more about who can prove that every automated purchase was discoverable, authorized, payable, defensible and reversible when something goes wrong.

Methodology and limitations

This article compares publicly announced products and company disclosures available as of October 7, 2026. “Best positioned” is an analytical assessment of product breadth, merchant integration, authorization primitives, fraud/identity capabilities and commercial distribution—not a claim that one processor has superior live agentic-commerce approval or fraud rates.

Agentic commerce is still early. Many deployments remain pilots or limited by geography. Published company claims describe capabilities, partnerships and selected performance metrics but do not provide a standardized third-party benchmark for agent-initiated payment conversion, fraud losses or chargeback rates. The comparison therefore gives more weight to verifiable product architecture and existing payments scale than to marketing forecasts.

The scale thought experiment applies 5%, 10% and 20% penetration to reported 2025 volume solely to show magnitude. Stripe, Adyen and Checkout.com define reported volume differently and use different currencies, so those figures should not be interpreted as directly comparable market shares.

Sources

1. Stripe — Agentic Commerce Suite

2. Stripe — Introducing our agentic commerce solutions / Shared Payment Tokens

3. Stripe — Agentic Commerce Suite launch

4. Stripe — Sessions 2026: economic infrastructure for AI

5. Stripe — Supporting additional payment methods for agentic commerce

6. Stripe — Agentic commerce getting-started guide

7. Stripe — 2025 annual letter

8. Adyen — Adyen Agentic announcement

9. Adyen — 2026 Fraud Report: Fraud’s identity crisis

10. Adyen — Agentic commerce and control

11. Adyen — H2 / FY 2025 financial results

12. Adyen — H1 2026 financial results

13. Checkout.com — Agentic commerce: your questions answered

14. Checkout.com — ACP adoption

15. Checkout.com — Chargebacks in agentic commerce

16. Checkout.com — 2025 volume / profitability

17. Checkout.com — 2026 annualised revenue / expected TPV

18. Checkout.com — Agentic Commerce 2026 report

19. Checkout.com — Network tokens

20. Checkout.com — Disputes

Johan Shamshad

Financial Markets Analyst and Journalist at Dave Finances |  More PostsBio ⮌

Johan Shamshad is a financial markets writer at Dave Finances covering cryptocurrencies, trading platforms, brokers, fintech, financial regulation, and developments across global markets. He previously worked at Gulf News, adding newsroom experience to his coverage of fast-moving financial and digital-asset markets.

His work focuses on identifying market-moving events, company developments, regulatory changes, product launches, and shifts in trading and financial infrastructure.

Johan contributes news and analysis designed to help readers understand not only what happened, but why a development matters and how it may affect the wider financial landscape.

• Johan Shamshad

  Visa vs Mastercard vs Stripe: Three Very Different Bets on Stablecoins
• Johan Shamshad

  Tokenized Stocks vs Real Stocks vs CFDs: What Do Retail Traders Actually Own?
• Johan Shamshad

  Robinhood vs Binance vs Kraken: Who Is Building the Real Everything App?
• Johan Shamshad

  Polymarket vs Kalshi vs Robinhood: Which Prediction Market Works Best for Traders?