Agentic Payments Statistics 2026
By Axis Intelligence Research
Co-author: Sarah Davis, Digital Finance & Fintech | Last updated: September 4, 2026 | License: CC BY 4.0
Nine markets have now recorded a publicly confirmed, fully authenticated payment initiated by an AI agent on a global card network, the first on January 28, 2026 and the most recent on July 2, 2026. Across those markets and the network readiness programs behind them, Axis Intelligence Research can name 39 institution-market pairs against a disclosed enrolment floor of 100 partners. The rails work. The disclosure does not.
Quick Answer: What Are the Key Agentic Payments Statistics for 2026?
According to Axis Intelligence Research, nine markets had a primary-source-confirmed live agent-initiated card payment as of September 4, 2026, and the global APRI (Agentic Payment Rail Index) reading stands at 60.3 out of 100. Asia Pacific leads at 80.0; North America, home to every major agentic payment protocol, scores 47.5. At least 100 partners are enrolled in network issuer-readiness programs, but only 39 institution-market pairs have been publicly named.
Key Findings
- According to Axis Intelligence Research, nine markets recorded a publicly confirmed authenticated agent-initiated card payment between January 28, 2026 and July 2, 2026: Australia, New Zealand, India, Singapore, Malaysia, South Korea, Hong Kong, Germany and the United States. (Axis Intelligence Research go-live registry, September 4, 2026)
- Axis Intelligence Research finds that the agentic payment institutional disclosure ratio is 39.0%: 39 named institution-market pairs against a disclosed enrolment floor of 100 partners in network issuer-readiness programs. Sixty-one enrolled participants remain unnamed in public materials. (Axis Intelligence Research calculation from Visa and Mastercard releases, September 4, 2026)
- Axis Intelligence Research finds Asia Pacific is the most agent-ready payments region in the world, with an APRI reading of 80.0, and North America the least ready of the four regions with a confirmed live transaction, at 47.5. (APRI, Axis Intelligence Research, as of September 4, 2026)
- Seventeen named issuers and processors executed live agentic card payments across Latin America and the Caribbean on a single announcement date, the largest named institutional cohort in any market to date. (Mastercard, March 24, 2026)
- The UK Financial Conduct Authority became the first financial regulator to name agentic payments in a published priorities document, committing to consider whether regulatory change is needed to support agentic AI payments. (FCA Regulatory Priorities: Payments, March 25, 2026)
The APRI: Axis Agentic Payment Rail Index
Every published number in this field measures demand: how many consumers say they would let an agent buy something, how much money analysts think will flow. Nobody measures the supply side, which is the part that actually gates the market. An agent cannot pay if no issuer will authorize the transaction.
The APRI (Agentic Payment Rail Index) is an Axis Intelligence Research metric scoring how far the payment acceptance layer in a region has moved from announcement to production. It runs 0 to 100. Every input is a fetched primary source, and every input is a row in the downloadable dataset.
Formula and weights
APRI = L + D + P + R
L Live execution (35 points, binary)
35 if at least one completed, authenticated, agent-initiated card payment
in the region is confirmed by a network, issuer or processor press release.
D Named-institution depth (30 points)
min(named institution-market pairs / 30, 1.0) x 30
Counts distinct financial institutions and processors named in primary
sources as participants in a live or enrolled agentic payment flow,
counted once per market.
P Issuer-readiness program (25 points)
25 network issuer-readiness program confirmed in-region with a disclosed
participant count or enablement figure
12.5 region covered by a network program with no disclosed regional count
0 no program confirmed
R Regulatory engagement (10 points, binary)
10 if a financial regulator in the region has named agentic payments in a
published policy, priorities or consultation document.
APRI regional readings, as of September 4, 2026
| Region | L | D | P | R | APRI |
|---|---|---|---|---|---|
| Asia Pacific | 35 | 20.0 | 25.0 | 0 | 80.0 |
| Latin America & Caribbean | 35 | 17.0 | 25.0 | 0 | 77.0 |
| Europe & UK | 35 | 2.0 | 25.0 | 10 | 72.0 |
| North America | 35 | 0.0 | 12.5 | 0 | 47.5 |
| CEMEA | 0 | 0.0 | 25.0 | 0 | 25.0 |
| Global (unweighted mean) | 60.3 |
Source: Axis Intelligence Research. Component inputs and their source URLs are in the CSV dataset. Worked example, Asia Pacific: 35 + (20/30 x 30 = 20.0) + 25 + 0 = 80.0. Global reading is the unweighted mean of the five regional readings.
What the index does and does not say. A high APRI means the acceptance side is demonstrably working in production and that the participants are willing to be named. It says nothing about transaction volume, which no network has disclosed, and nothing about consumer willingness to delegate, which is measured separately in our agentic commerce statistics analysis. APRI is a supply-side reading. It is designed to move as issuers go live, and it will be recalculated each quarter.
Sarah Davis: The number that should stop a payments executive mid-sentence is North America at 47.5. ACP came out of San Francisco. AP2 came out of Mountain View. Trusted Agent Protocol came out of Foster City. Every protocol governing how an agent proves it has permission to spend your money was written within a ninety-minute drive of the same airport, and yet the market where you can point at a named US bank that has authorized one of these transactions in production does not exist in the public record. The specification-writing is American. The going-live is happening in Kuala Lumpur, Auckland and Bogota.
Which Markets Have Completed a Live Agentic Payment?
This registry is the compiled output of every primary network, issuer and processor announcement Axis Intelligence Research could locate and verify. A market enters only when a press release confirms a completed, authenticated, agent-initiated payment, not a demonstration of intent or a stated roadmap.
Agentic payment go-live registry
| Date | Market | Named financial institutions | Rail / program | Source |
|---|---|---|---|---|
| 2026-01-28 | Australia | Commonwealth Bank of Australia; Westpac | Mastercard Agent Pay | Mastercard |
| 2026-02-17 | New Zealand | Westpac New Zealand | Mastercard Agent Pay | Mastercard |
| February 2026 | India | Axis Bank; RBL Bank | Mastercard Agent Pay Framework | Mastercard |
| 2026-03-04 | Singapore | DBS Bank; UOB | Mastercard Agent Pay | Mastercard |
| 2026-03-04 | Malaysia | CIMB Group; RHB Banking Group | Mastercard Agent Pay | Mastercard |
| 2026-03-17 | South Korea | None named | Mastercard Agent Pay | Mastercard |
| 2026-03-24 | Latin America & Caribbean | 17 issuers and processors (listed below) | Mastercard Agent Pay | Mastercard |
| 2026-03-27 | Hong Kong | HSBC; DBS Hong Kong; Citi Hong Kong; Hang Seng Bank; Standard Chartered Hong Kong; Mox Bank | Mastercard Agent Pay | Mastercard |
| 2026-07-02 | Germany | ING (issuer); Worldline (processor) | Visa Intelligent Commerce | Worldline |
| 2025 | United States | None named | Mastercard Agent Pay; Visa Intelligent Commerce | Mastercard; Visa |
Source: Axis Intelligence Research compilation from primary network, issuer and processor press releases. Every source URL and retrieval date is in the CSV dataset. Taiwan is excluded pending verification of a source in English.
The seventeen Latin American and Caribbean participants named by Mastercard on March 24, 2026 were BAC, Banco de la Nacion Argentina, Banco Falabella, Banco Galicia, Banco Itau, Banco Ripley, Banco Security de BICE, Bancolombia, Banamex, Cencosud-Scotiabank, Davivienda, Dock, Evertec, Oriental Bank, Pomelo, Santander and ueno bank. Their transactions covered debit and credit cards buying cosmetics, groceries, books and digital goods, and Mastercard states that nearly all issuers in the region are already enabled with its agentic token technology, which is why the region reached seventeen named participants in one announcement rather than two.
The European entry is instructive for a different reason. On July 2, 2026, Worldline, ING and Visa completed an agent-driven transaction in Germany under existing Strong Customer Authentication rules, with the consumer confirming intent through a Visa Payment Passkey and ING authorizing as issuer. Europe did not need a regulatory carve-out to run the transaction. It ran inside PSD2 as written.
Sarah Davis: Notice what the registry is really a registry of. Nine markets, and in seven of them the merchant was a taxi, a cinema, a hotel or a grocery basket. These are card-not-present transactions with a ticket size somewhere between a coffee and a weekend. Nobody is agent-authorizing a mortgage. The technology being proved out is not autonomous spending. It is delegated, pre-scoped, biometrically confirmed spending on small amounts, which is a far narrower thing and a far more sellable one.
How Ready Are Issuers for Agent-Initiated Transactions?
Live transactions are proofs. Issuer enrolment is the pipeline, and it is where the volume will eventually come from. Both global networks now run formal programs that put the issuer, not the merchant, at the centre.
Network issuer-readiness programs
| Program | Network | Region | Disclosed participants | Announced | Source |
|---|---|---|---|---|---|
| Agentic Ready | Visa | Europe & UK | 20+ partners | 2026-03-17 | Visa |
| Agentic Ready | Visa | Asia Pacific | 50+ partners across 10 markets | 2026-04-30 | Visa |
| Agentic Ready | Visa | CEMEA | 30+ issuers | 2026-05-20 | Visa |
| Intelligent Commerce | Visa | Global | 100+ ecosystem partners; 30+ in sandbox; 20+ agents integrating | 2025-12-18 | Visa |
| Intelligent Commerce Connect | Visa | Global | 4 agent protocols accepted | 2026-04-08 | Visa |
| Agent Pay | Mastercard | Latin America & Caribbean | Nearly all regional issuers enabled with agentic tokens | 2026-03-24 | Mastercard |
Source: Axis Intelligence Research compilation from Visa and Mastercard primary releases. Where a publisher states “over” or “more than” a figure, the floor value is used.
Visa Agentic Ready is explicitly an issuer-side program: participants test card enrolment, tokenization, authentication and authorization for agent-initiated flows in production environments before the volume arrives. In Vietnam alone, Visa named five participating issuers: Asia Commercial Bank, MB, Sacombank, Techcombank and VPBank. That single market disclosed more named issuers than the entirety of North America and Europe combined.
The disclosure gap
Axis Intelligence Research computes the agentic payment institutional disclosure ratio by dividing the number of institution-market pairs named in primary sources by the disclosed floor of enrolled program participants:
Named institution-market pairs = 20 (Asia Pacific) + 17 (LAC) + 2 (Europe) = 39
Disclosed enrolment floor = 50 (APAC) + 30 (CEMEA) + 20 (Europe/UK) = 100
Disclosure ratio = 39 / 100 = 39.0%
Sixty-one enrolled participants are, in public terms, anonymous. For a journalist trying to establish whether a specific bank supports agent-initiated payments, or for an enterprise buyer trying to establish whether their acquirer does, that is the operative constraint on the whole category. The two figures are not perfectly commensurate (a named live-transaction participant is not identical to an enrolled test participant), which is exactly why we publish both numbers rather than merging them into one.
Institutional sentiment against institutional action
The Cloud Security Alliance and Anjuna State of Cloud and AI for Financial Services 2026 survey puts the belief gap in numbers: 85% of financial services respondents expect AI agents to initiate and execute payment transactions on behalf of consumers, and 65% believe this requires an entirely new authorization model. Meanwhile 62% of respondent organisations already use AI agents somewhere, but only 5% of those deployments operate at high autonomy.
| Indicator | Value | Source |
|---|---|---|
| Expect agents will initiate and execute payments | 85% | Cloud Security Alliance / Anjuna, June 2026 |
| Believe a new authorization model is required | 65% | Cloud Security Alliance / Anjuna, June 2026 |
| Organisations already using AI agents | 62% | Cloud Security Alliance / Anjuna, June 2026 |
| Agent deployments at high autonomy | 5% | Cloud Security Alliance / Anjuna, June 2026 |
| US shoppers using AI for at least one shopping task | 47% | Visa / Morning Consult, October 2025 |
| Australians who have used AI assistants to shop | 48% | PayPal Australia / YouGov, cited January 2026 |
Sarah Davis: Eighty-five percent expect it, five percent do it at high autonomy. That is not hypocrisy, it is a control problem with a specific shape. An issuer’s risk model is built around a decisioning window measured in milliseconds and a fraud liability regime built around a human being present at the moment of consent. Take the human out of that moment and the model still runs, but the liability question underneath it has no settled answer. The 65% who say the authorization model has to change are the ones who have read their own chargeback rules recently.
Which Protocols Actually Move the Money?
The word “protocol” is doing several different jobs in this market. Four specifications launched between September 2025 and January 2026, and they are not competitors so much as different floors of the same building.
| Protocol | Originator | Released | Layer it governs | Payment role |
|---|---|---|---|---|
| Agentic Commerce Protocol (ACP) | Stripe and OpenAI | 2025-09-29 | Checkout session between agent and merchant | Passes a Shared Payment Token; merchant stays merchant of record |
| Agent Payments Protocol (AP2) | Google, 60+ organisations | 2025-09-16 | Authorization evidence | Three signed mandates: Intent, Cart, Payment |
| Trusted Agent Protocol (TAP) | Visa, 10+ partners | 2025-10-14 | Agent identity at the merchant edge | Distinguishes a legitimate agent from a malicious bot |
| Agent Pay / Agentic Tokens | Mastercard | 2025-04-29 | Payment credential | Network token issued per agent, with issuer visibility |
Source: Axis Intelligence Research compilation from Stripe, Google Cloud, Visa and Mastercard primary announcements.
Google’s AP2 launched with more than 60 collaborating organisations including Adyen, American Express, Ant International, Coinbase, Etsy, Intuit, JCB, Mastercard, PayPal, Revolut, Salesforce, ServiceNow, UnionPay International and Worldpay. Its contribution is narrow and important: it separates the question of whether the user authorized this purchase from the question of whether the payment is valid. The first is answered by a chain of cryptographically signed mandates. The second is answered by whatever rail moves the money, card, bank transfer or stablecoin.
The convergence is already visible in the plumbing. On March 3, 2026, Stripe extended Shared Payment Tokens to carry both Mastercard Agent Pay and Visa Intelligent Commerce agentic network tokens, plus Affirm and Klarna for buy now, pay later. Five weeks later Visa announced Intelligent Commerce Connect, which accepts agent-initiated payments across four protocols at once: Trusted Agent Protocol, Machine Payments Protocol, ACP and UCP. Neither company is picking a winner. Both are building adapters, which tells you what they expect the next two years to look like.
Sarah Davis: Follow the token and the business model appears. An agentic token is a network token with a policy attached, and network tokens are the highest-margin thing the card networks sell. Every agent-initiated transaction that routes through Agent Pay or Intelligent Commerce is a transaction that stays on the card rail rather than migrating to account-to-account or a stablecoin. The protocols are being given away for free because the credential is the product, and the credential is where the take rate lives.
How Are Regulators Treating Agent-Initiated Payments?
The UK moved first among major jurisdictions. In its March 2026 Regulatory Priorities: Payments report, the Financial Conduct Authority committed to modernising payments regulation, including “whether change or development of regulation is needed to support agentic AI payments.” According to Axis Intelligence Research, that is the first instance of a financial regulator naming agentic payments as a live policy question inside a formal priorities document rather than a speech or a sandbox announcement.
The reason it matters is structural rather than rhetorical. UK payment services rules require the payer to consent to a specific transaction. An agent that selects the transaction breaks the assumption the rule was drafted around. The same report shows the FCA is working from a position of scale rather than experiment: over 16 million people and businesses used open banking in the UK during 2025, generating over 29 million payments per month on average. A regulator supervising that volume of programmable payments has an obvious interest in getting agent consent right before it is retrofitted.
Everywhere else, the position as of September 2026 is that existing frameworks apply. The German transaction demonstrated that Strong Customer Authentication and agent initiation are compatible today, which reduces the urgency for European rulemaking and correspondingly raises the value of the passkey and mandate infrastructure that makes the compatibility work. Our EU AI Act compliance tracker and AI regulation tracker follow the adjacent obligations as they take effect.
Where Is the Fraud and Liability Exposure in Agentic Payments?
The agent-initiated flow introduces one genuinely new artifact: a machine-readable record of what the consumer authorized, generated before the purchase rather than reconstructed after a dispute. Mastercard’s Verifiable Intent and AP2’s mandate chain are both attempts to make that record tamper-resistant and portable across issuers.
That is a meaningful improvement over the alternative, which is what most agent purchasing looked like before these protocols existed: browser automation filling out a checkout form with a stored card, indistinguishable at the issuer from a bot attack. Distinguishing the two is the entire purpose of Visa’s Trusted Agent Protocol at the merchant edge, and it is why the networks are insisting on agent registration before any transaction crosses their rails.
Two unresolved items sit underneath all of it. The first is scope enforcement: if a consumer authorizes camping supplies and the agent buys a tent plus a television, whose control failed, and at which layer. The second is dispute rights: a chargeback framework built for a human who can testify about what they intended now has to accommodate a signed mandate as the evidence of intent. Neither has been settled by a regulator or a network rule change that Axis Intelligence Research could locate as of September 2026, and we would rather publish that gap than fill it with a projection. Adjacent exposure patterns are tracked in our AI agent security incident tracker and quantified for the broader loss picture in cybercrime statistics.
What Happens to Agentic Payments Next?
Three observable things, each with a date attached.
The disclosure ratio is the first. If it stays near 39% into 2027, it tells you issuers are enrolling defensively rather than commercially, keeping optionality without wanting the marketing risk. If named participants climb toward the enrolment floor, the category has crossed from pilot into product.
North America is the second. The APRI reading of 47.5 is not a technology gap. It reflects that no US or Canadian issuer has been publicly named in a completed agent-initiated transaction in the sources we verified. The first named US bank in a production agentic flow will move that reading by up to 30 points on its own.
Settlement rail diversification is the third. AP2 was payment-agnostic from day one, and Stripe has already widened the agentic credential to include buy now, pay later alongside card. According to Axis Intelligence Research, the earliest meaningful signal of rail migration will be an agent-initiated account-to-account or stablecoin payment announced by an issuer rather than by a protocol author. Our stablecoin statistics dataset tracks the settlement side of that question.
Methodology
Collection
Axis Intelligence Research built this dataset by fetching and reading primary announcements from payment networks (Visa, Mastercard), processors (Stripe, Worldline), protocol authors (Google Cloud), a financial regulator (UK Financial Conduct Authority) and one industry survey publisher (Cloud Security Alliance with Anjuna) between September 2 and September 4, 2026. Every quantitative claim in this article corresponds to a row in the downloadable CSV carrying the source organisation, document title, URL and retrieval date.
A market enters the go-live registry only when a primary source states that a completed, authenticated, agent-initiated payment took place. Statements of intent, roadmap announcements and demonstrations without a confirmed authorization are excluded. Where a publisher writes “more than 50 partners,” the dataset records 50 and flags the figure as a floor, because inventing a midpoint would create false precision.
APRI construction
The formula, weights and worked arithmetic appear in the index section above and in the method_note column of every axis_calculated = yes row. The index is reproducible: a reader with the same CSV can recompute all six readings.
Three scoring decisions are worth stating plainly. Named-institution depth is capped at 30 pairs so that one unusually generous regional announcement does not dominate the reading. The issuer-program component awards half credit where a network operates in a region without disclosing a participant count, because presence and transparency are different things. And CEMEA scores zero on live execution because Axis Intelligence Research could not confirm a completed agent-initiated transaction in the region from a primary source, despite Visa disclosing 30-plus enrolled issuers there. We have recorded that as a finding rather than resolving it by inference.
What this dataset does not measure
There is no public agentic payment transaction volume. No network, issuer or processor has disclosed a count or a value of agent-initiated transactions, and Axis Intelligence Research declined to estimate one. Any figure circulating for the size of the agentic payments market in 2026 is a forecast of a market whose current volume is undisclosed, and should be read accordingly. Consumer demand-side measures, market sizing and merchant readiness are covered separately in our agentic commerce statistics analysis to keep the supply-side and demand-side pictures methodologically distinct.
About This Dataset
Dataset title: Agentic Payments Statistics 2026
Creator and publisher: Axis Intelligence Research
Contents: 53 observations covering the agentic payment go-live registry, network issuer-readiness program enrolment, protocol specifications, regulatory milestones, institutional survey readings and six APRI index values
Geographic coverage: Global, with Asia Pacific, Latin America and the Caribbean, Europe and the UK, North America and CEMEA breakdowns, plus 12 individual markets
Temporal coverage: April 2025 to September 2026
File: agentic-payments-statistics-2026.csv (14 columns, UTF-8, comma-separated)
License: CC BY 4.0. Free to use, share and adapt with attribution to Axis Intelligence Research.
Citation line: Axis Intelligence Research, Agentic Payments Statistics 2026
Mirrors: Hugging Face, Kaggle and GitHub on publication
Cite This Research
APA: Axis Intelligence Research, & Davis, S. (2026, September 4). Agentic payments statistics 2026: Live markets, issuer readiness, protocols and the APRI regional index. Axis Intelligence. https://axis-intelligence.com/agentic-payments-statistics/
MLA: Axis Intelligence Research, and Sarah Davis. “Agentic Payments Statistics 2026: Live Markets, Issuer Readiness, Protocols and the APRI Regional Index.” Axis Intelligence, 4 Sept. 2026, axis-intelligence.com/agentic-payments-statistics/.
Chicago: Axis Intelligence Research and Sarah Davis. “Agentic Payments Statistics 2026: Live Markets, Issuer Readiness, Protocols and the APRI Regional Index.” Axis Intelligence. September 4, 2026. https://axis-intelligence.com/agentic-payments-statistics/
Embeddable citation: According to Axis Intelligence Research, nine markets recorded a publicly confirmed live agent-initiated card payment between January and July 2026, and the global APRI (Agentic Payment Rail Index) reading is 60.3 out of 100 as of September 4, 2026, with Asia Pacific at 80.0 and North America at 47.5. (Axis Intelligence Research, September 2026, CC BY 4.0)
Frequently Asked Questions
Which banks have actually authorized an AI agent to make a payment?
Thirty-two distinct institutions and processors have been named in primary sources as participants in live agentic card payments. In Asia Pacific, fourteen: HSBC, DBS (Hong Kong and Singapore), Citi Hong Kong, Hang Seng Bank, Standard Chartered Hong Kong, Mox Bank, UOB, CIMB, RHB, Axis Bank, RBL Bank, Commonwealth Bank of Australia, Westpac and Westpac New Zealand. In Latin America and the Caribbean, seventeen issuers and processors including Santander, Bancolombia, Banamex, Banco Itau and Banco Galicia. In Europe, one issuer, ING, alongside processor Worldline. A further five Vietnamese issuers are named as enrolled in Visa Agentic Ready rather than as live-transaction participants. No US or Canadian issuer had been publicly named as of September 4, 2026.
Does an agent-initiated transaction satisfy Strong Customer Authentication?
It can, and one has. In the July 2, 2026 German transaction, the consumer confirmed purchase intent through a Visa Payment Passkey using biometric authentication, and ING authorized as issuer, entirely within existing SCA requirements. The pattern that makes this work is that the agent selects, and the human confirms with a possession-and-inherence factor. A fully unattended agent purchase with no confirmation step is a different regulatory question and has not been demonstrated under SCA.
What is an agentic token, and how is it different from an Apple Pay token?
Both are network tokens: a surrogate credential that replaces the card number so the real PAN is never exposed. The difference is what is bound to the token. A wallet token assumes a human is present and authenticating at the moment of payment. An agentic token is issued uniquely to a specific agent and carries the scope of what that agent is permitted to do, so the issuer can see that an agent initiated the transaction and can decline activity outside the authorized scope. Mastercard calls these Agentic Tokens; Visa and Stripe refer to agentic network tokens.
Who is liable if an AI agent buys the wrong thing?
There is no settled answer as of September 2026, and Axis Intelligence Research declines to assert one. What has changed is the evidence available: AP2’s Intent, Cart and Payment mandates and Mastercard’s Verifiable Intent both create a signed, timestamped record of what the consumer authorized before the purchase, which is materially better than reconstructing intent after a dispute. Whether that record shifts liability between issuer, agent provider and merchant will be determined by network rules and regulators, not by protocol design. The FCA has flagged the question; no jurisdiction has answered it.
Can an issuer decline an agent-initiated transaction?
Yes, and that is central to how the networks have designed these programs. The authorization request carries additional data identifying the transaction as agent-initiated, which lets the issuer apply its own controls, decline outside authorized scope, or step up authentication. Visa’s Agentic Ready program exists specifically so issuers can test that decisioning in production before volume arrives, covering card enrolment, tokenization, authentication and authorization.
Do agentic payments run on their own rails, or on existing card rails?
Overwhelmingly on existing card rails today. Mastercard’s own framing of the Latin American deployment is that AI-initiated transactions moved from concept to execution without introducing new rails or credentials. AP2 is deliberately payment-agnostic and supports account-to-account transfers and stablecoins, and Stripe has extended the agentic credential to Affirm and Klarna, but every live transaction in the Axis go-live registry settled over a card network.
What is the difference between ACP, AP2, UCP and Trusted Agent Protocol?
They govern different layers and are frequently used together. ACP, from Stripe and OpenAI, handles the checkout session between agent and merchant. AP2, from Google and 60-plus partners, handles authorization evidence through signed mandates. Trusted Agent Protocol, from Visa, handles agent identity at the merchant edge so a legitimate agent is distinguishable from a malicious bot. UCP, from Google, covers the wider commerce flow from discovery to checkout. Visa Intelligent Commerce Connect accepts payments initiated through four of these at once, which is a fair signal that no single protocol is expected to win outright.
How many agent-initiated payments have actually happened?
Nobody has said. No network, issuer or processor has published a count or a value of agent-initiated transactions. Visa described hundreds of controlled agent-initiated transactions completed with partners as of December 2025; that is the largest volume figure Axis Intelligence Research located from a primary source, and it is an order of magnitude, not a metric. Any market-size figure circulating for agentic payments in 2026 is a forecast built on an undisclosed base.
Why does Asia Pacific score higher than the United States on agentic payment readiness?
Because APRI measures deployment and disclosure, not invention. Asia Pacific has eight markets with confirmed live transactions, twenty named institutions and a disclosed 50-plus-partner issuer program. North America has confirmed live transactions but no publicly named issuer participants and no disclosed regional program enrolment count. The protocols were largely written in the United States; the production evidence has been published elsewhere.
What would move the APRI reading in the next quarter?
Four things, in descending order of impact. A named US or Canadian issuer completing a production agentic transaction would add up to 30 points to the North American reading. A confirmed live transaction in CEMEA would add 35 points there. A financial regulator outside the UK naming agentic payments in a published document would add 10 points to that region. And additional named institutions in Europe would lift the depth component, which currently sits at 2.0 out of 30.
Related Axis Intelligence Research: Agentic Commerce Statistics 2026 · Stablecoin Statistics 2026 · Cybercrime Statistics 2026 · AI Agent Security Incident Tracker · EU AI Act Compliance Tracker · AI Regulation Tracker · Digital Finance statistics