AI agents have made more than 138 million crypto payments, yet only 6.8% of US shoppers have let one check out for them. This stat hub pulls together on-chain data, card network rollouts, trust surveys and 2030 forecasts, and explains which numbers you can rely on.

Table of Content

Most Read

Xiaomi 18 Pro vs POCO F8 Pro: Is the Chinese Flagship Worth $300 More?

  • Nov 12
  • 6 mins read

Poco C85 5G vs Realme C100i: The Older Phone Wins on Price, Answered Question by Question

  • Nov 14
  • 9 mins read

The Oppo A7 Ultra Says 12,000mAh on the Box. Its Paperwork Says 11,640mAh. Both Are True

  • Nov 14
  • 6 mins read

Realme GT 7T vs Moto G77: The Realme Wins 5 of 7 Rounds, but the Moto Costs Half as Much

  • Nov 12
  • 10 mins read

vivo X300 Pro vs Galaxy S25 Ultra 2026: Which Flagship Makes More Sense?

  • Nov 10
  • 6 mins read

Sponsored

TL;DR

  • AI agents made about 138.5 million x402 payments on Base by Q3 2026, up from 58,800 a year earlier.
  • More than 87% of those payments cost between $0.0001 and $0.10, far below what card fees can handle.
  • TRM Labs estimates only 0.6% to 7.5% of filtered x402 value comes from real AI agents.
  • Just 38% of Americans are comfortable letting AI complete a purchase, and 60% want to approve every spend.
  • Forecasts for 2030 range from $300 billion in US agentic sales to $5 trillion in global agentic commerce.

On October 14, 2026, Chainalysis reported that AI agents have made more than 138 million crypto payments. Most of them cost less than a dime. At the same time, Stripe said agent purchases on its Link wallet jumped 38-fold. Even so, recent surveys show most people still won’t let an AI spend their money unsupervised. For this reason, we pulled together the most reliable AI agents in payments statistics as of October 24, 2026. Below, you’ll find data on crypto rails, card networks, consumer trust, fraud and forecasts, plus notes on where the numbers conflict.

AI Agent Payments at a Glance

Table 1: AI Agent Payments at a Glance (October 2026)

AI Agent Payments at a Glance (October 2026)

Note: “x402” is Coinbase’s payment standard for AI agents, explained below. First Page Sage’s figures come from its own shopper panel and partner data, which hasn’t been published in full.

How to Read These Numbers

AI agent payment statistics are easy to quote and easy to misread. An AI agent is software that takes actions for you. For example, it can compare prices, book a table or pay for a service. However, “agent payments” can mean three different things:

  • Agent-assisted shopping: the AI helps you research, but you still click “buy.” For example, most survey and traffic data measures this.
  • Agent-completed checkout: the AI fills the cart and pays, usually after you approve. That’s also what Visa, Mastercard and Stripe are building for.
  • Machine-to-machine payments: software pays other software, often a fraction of a cent. In contrast, most on-chain data measures this.

This difference matters more than any single number. For instance, holiday sales reports describe the first group. Crypto transaction counts, however, describe the third. Meanwhile, the middle group, where an agent spends your money in a store, is still the smallest.

To make this clear, every statistic below is labeled by evidence type. These are on-chain data, company disclosures, independent surveys, vendor surveys and forecasts. In particular, vendor surveys come from companies that sell agent payment products. As a result, treat them and forecasts as direction, not settled fact.

Agentic Commerce Market Size & Forecasts

Global Forecasts for 2030

Agentic commerce means shopping where an AI agent finds, picks, or buys products for you. So far, three global forecasts stand out. First, Juniper Research projects agentic commerce spend will reach $1.5 trillion in 2030. Juniper also names trust as the biggest barrier. In contrast, McKinsey’s October 2025 estimate is far larger, at $3 trillion to $5 trillion. Meanwhile, Edgar, Dunn & Company sits in between at $1.7 trillion, a figure PayPal cited in April 2026.

US Forecasts for 2030

US forecasts, on the other hand, fall into two groups. For example, Bain projects $300 billion to $500 billion, or roughly 15% to 25% of US e-commerce. Similarly, Morgan Stanley expects up to $385 billion, or 10% to 20% of online spending. McKinsey, however, runs far higher. It expects $900 billion to $1 trillion in agent-orchestrated US retail revenue, for goods only.

Figure 1: 2030 Agentic Commerce Forecasts by Firm and Scope


2030 agentic commerce forecasts from Morgan Stanley, Bain, McKinsey, Juniper Research and Edgar, Dunn & Co.

Why the Forecasts Disagree

Much of the spread comes down to the word orchestrated. McKinsey’s US figure appears to count sales an agent shapes along the way, while Morgan Stanley and Bain stay closer to purchases agents actually make. Even among global estimates, McKinsey’s high end is 3.3x Juniper’s.

Timing matters too. Five of the six forecasts were published before OpenAI pulled its in-chat checkout in March 2026, the clearest real-world test of agent buying so far. Only Juniper’s came after, and it’s also the most cautious about trust. That doesn’t make the older numbers wrong, but it does suggest they were drawn before the market showed how slowly people hand over their cards.

A rough reality check helps. Agent-completed orders made up 0.68% of US e-commerce orders in Q3 2026. Reaching Bain’s 15% to 25% range by 2030 would take something like a 22x to 37x increase in four years. That comparison isn’t exact, since one measures orders and the other measures spending, but the gap shows how much of these forecasts rests on adoption that hasn’t happened yet.

AI Agents Are Already Shaping Online Spending

Holiday 2025: AI Traffic and AI-Influenced Sales

Agents’ influence on shopping showed up clearly in the 2025 holiday season. Adobe found that traffic to US retail sites from generative AI tools rose 693.4% year over year between November 1 and December 31, 2025, while US online spending hit a record $257.8 billion (+6.8%). Salesforce reported that AI and agents drove 20% of global holiday retail sales, about $262 billion, out of a record $1.29 trillion. Shoppers referred from AI search converted 9x more often than those from social media, and retailers running their own AI agents grew sales 6.2% versus 3.9% for those that didn’t.

These are big numbers, and they’re the ones most often attached to the phrase “agentic commerce.” But they measure influence, not payment. A shopper who asked ChatGPT for gift ideas and then bought on a retailer’s site counts here. The 693% jump also starts from a small base, which Adobe itself points out. What the data does show is that chatbots are becoming a real front door to online stores, and that tends to come before people trust the same tools to pay.

Agent-Completed Checkout in 2026

Completed agent purchases are much smaller but growing fast. First Page Sage’s study, covering July 13 to October 16, 2026, found that 61.4% of US online shoppers use AI for product research, but only 6.8% have let an agent complete checkout with their approval. Agent-completed orders rose from 0.27% to 0.68% of US e-commerce orders between Q2 and Q3 2026, and the study credits 71% of that jump to Meta’s Muse, which launched September 8.

Stripe’s numbers point the same way. According to data Stripe shared with Upstarts Media on October 18, agentic transactions on its Link wallet rose 38x in September 2026, and payment volume rose 20x, with an average purchase of about $100. Both data sets peak in the same month for the same reason: a consumer app with a huge built-in audience made agent buying a default feature, and Meta’s Muse hit 2.5 million downloads while Amazon blocked it. The weak spot is that neither Stripe nor First Page Sage has published the full underlying data, so the growth rates are more reliable than the totals.

What OpenAI’s Instant Checkout Taught the Market

The clearest failure so far came from the biggest AI brand. OpenAI launched Instant Checkout inside ChatGPT on September 29, 2025, using the Agentic Commerce Protocol (ACP) it built with Stripe. On March 5, 2026, OpenAI said it would end the feature because it “wasn’t generating conversions.” Reports put first-month trial at just 8% of US adult ChatGPT users, with only about a dozen Shopify merchants integrating it.

This may be the most useful single data point in the whole category. People were happy to research inside a chatbot, but they still wanted to pay on a site they knew, with saved promo codes and familiar return policies. Muse’s early success six months later suggests the answer wasn’t “people won’t let AI buy,” but “people need a reason and a familiar app to try it.”

On-Chain Agent Payments: x402 Statistics

Transaction Growth

x402 is a payment standard Coinbase published in May 2025. It revives an old, mostly unused web code, “HTTP 402 Payment Required,” so a website can quote a price and an AI agent can pay instantly with a stablecoin, a cryptocurrency pegged to the US dollar. Because these payments land on public blockchains, they’re the easiest agent payments to count.

Chainalysis reports about 138.5 million cumulative x402 transactions on Base, Coinbase’s low-fee blockchain network, by the end of Q3 2026. A year earlier, the count stood at about 58,800, roughly 2,355x growth. Separately, x402.org data reported in mid-October showed 43,100 monthly active senders over 30 days, up 152%, and 75.41 million transactions in that window.

Put those two figures side by side and an imbalance appears: about 75 million payments from 43,100 senders works out to roughly 1,750 payments per wallet per month. No person pays that often. Either a small number of automated programs are doing most of the work, or a handful of very busy wallets are inflating the count, and the data that follows suggests it’s a mix of both.

Most Agent Payments Cost Less Than a Dime

More than 87% of x402 payments in Chainalysis’ review were between $0.0001 and $0.10, with $0.01 to $0.10 the most common range. Card networks can’t process payments that small profitably, which is the main reason agents use crypto for this job at all.

Figure 2: Average Value per AI Agent Payment, by Dataset


The chart shows two different economies sharing one label. On crypto rails, the average agent payment ranges from 2 cents to 27 cents. On Stripe’s Link wallet, where agents buy retail goods with cards, the average is about $100, roughly 370x larger. When someone says “AI agents made X payments,” it’s worth asking which of these two worlds they mean. The BeInCrypto sample is the most extreme: 90.8% of its payments were under one cent, and the most-used paid service, an AI-generated crypto safety check, cost two cents a request. That looks less like shopping and more like a meter running on API calls.

Agent Wallets Are Getting Stickier

Wallet behavior is where the data looks most like real automation. Chainalysis counted only wallets that paid on two or more separate days, to filter out one-time experiments.

Figure 3: How x402 Wallet Behavior Changed Between Cohorts


The 73-days-to-18-minutes shift is the most telling number in this report. People open a crypto wallet and then wait days or weeks before using it. A wallet that gets funded and starts paying within minutes behaves like software built for one job. The rise in repeat top-ups adds a second signal: these wallets are being refilled like a prepaid meter, not abandoned after a test. Neither number proves an AI is behind the wallet, since a simple script would act the same way, but it’s the strongest evidence so far that automated spending, rather than curious humans, is driving growth. AI agents’ crypto payments and what China’s 0.1% share tells us go deeper into the same report.

What AI Agents Are Buying

Early x402 activity was mostly fads. Memecoins such as PING drew waves of new wallets and then faded. Chainalysis says more lasting demand now sits in information services: code review, financial intelligence, and data and search APIs (application programming interfaces, the connections apps use to request data from each other). Model inference, paying an AI model to produce an answer, is one of the largest categories, which means AI agents are increasingly paying other AIs. Industry experts quoted in the report still describe the market as being in an “experimentation phase,” and the mix supports that: agents are mostly buying the inputs they need to work, not products a person would recognize as shopping.

Base vs Solana: Why the Chain Numbers Disagree

If you’ve seen headlines saying Base dominates agent payments and others saying Solana does, both can be right. They measure different things over different windows.

Table 2: Base vs Solana in x402 Payments, by Source and Metric

Base vs Solana in x402 Payments

Note: Base and Solana shares for x402.org were calculated by Memeburn from reported sender counts. Algorand had about 9,100 senders (~21%) in the same window.

The pattern is clear once the metrics are separated. Base leads on how many different wallets are paying, while Solana has led on how many payments happen during busy weeks. For judging real adoption, distinct senders is the more useful number, because it’s harder to inflate: one bot can send a million payments, but it still counts as one sender. Solana’s bursts show its low fees and speed suit high-frequency machine traffic, but they also show how a few very active wallets can rewrite a chain’s market share overnight. Algorand’s showing, with about a fifth of active senders, is the surprise most coverage misses.

How Much of It Is Real Agent Activity?

This is the question most stat pages skip, and it’s the one that matters most.

Figure 4: How Much x402 Value Comes From Real AI Agents


TRM Labs analyzed 198.9 million x402 settlements worth about $52.7 million since May 2025. Removing self-payments, flows dominated by one or two payers, and sellers with fewer than 10 distinct buyers cut the value roughly in half, to $25.62 million. Of that, TRM estimated only 0.6% to 7.5% looked like it came from AI agents. Applying that range to the filtered value gives roughly $150,000 to $1.9 million in agent spending across 16 months.

Two things stand out. First, half of the starting value looked like wash activity, such as wallets paying themselves, which is a reminder that raw x402 totals overstate the market. Second, the high end of TRM’s range is 12.5x the low end. That width says more about the state of measurement than about agents: on a public blockchain, a scheduled script and an AI agent look almost the same, and no one has a reliable way to tell them apart yet.

BeInCrypto’s readiness scores reach a similar verdict from a different angle. On a provisional 100-point scale, micropayments scored 60.7, trading and investment agents 39.1, and institutional workflows just 12.0. Of 19 AI trading and financial tools it reviewed, 14 operated under spending or trading limits set by people, and on the crypto exchange Hyperliquid, AI-labeled tools made up less than 0.55% of trading volume. The technology works best where the stakes are smallest, which is probably the right order for trust to build.

Stablecoins Are the Default Currency for Machine Payments

Almost every on-chain agent payment uses one coin. TRM Labs found that $52.47 million of $52.68 million in settled x402 value, or 99.6%, moved in USDC, the dollar stablecoin issued by Circle.

For scale, the total stablecoin market was worth about $300.7 billion as of October 24, 2026, with USDT at $182.5 billion (60.7%) and USDC at $71.6 billion (23.8%). All x402 value TRM tracked equals roughly 0.07% of USDC’s current supply. In other words, agents depend completely on stablecoins, but stablecoins don’t depend on agents at all, at least not yet. USDC’s near-monopoly here also flips the wider market, where Tether’s USDT is more than twice as large. Coinbase’s role in launching both x402 and Base, and its long partnership with Circle, explains much of that.

Big payment companies are moving in anyway. In June 2026, the Open USD (OUSD) consortium launched with Visa, Mastercard and Stripe as founding partners and more than 140 companies backing it. Circle describes the emerging split neatly: card rails for consumer-prompted agent payments, and stablecoin rails for machine-to-machine settlements too small for card economics. That division fits every data point above, from the $100 Link purchases to the two-cent API calls. If you’re new to holding stablecoins, our guide to the best crypto wallets for beginners covers the basics.

Card Networks and Fintech: Visa, Mastercard, Stripe and Google

Visa Intelligent Commerce

As of December 18, 2025, Visa said it was working with more than 100 partners on Visa Intelligent Commerce (VIC), its system that lets AI agents pay with Visa credentials. More than 30 partners were building in its sandbox, more than 20 agents were integrating directly, and “hundreds” of real-world agent-initiated transactions had been completed. Visa predicted millions of consumers would use AI agents to buy things by the 2026 holiday season.

Visa also introduced the Trusted Agent Protocol (TAP) in October 2025 with more than 10 partners, to help merchants tell legitimate AI agents from malicious bots. In April 2026, it launched Visa Intelligent Commerce Connect, a single integration that supports four agent protocols, and by April 29 its Agentic Ready program had more than 20 live partners in the UK and Europe, with a rollout planned to more than 85 partners in Asia Pacific and Latin America. Visa’s partner counts keep rising, but its last public transaction count was still “hundreds.” The upcoming holiday season is the first real test of its “millions of consumers” prediction.

Mastercard Agent Pay and Agent Pay for Machines

Mastercard launched Agent Pay in April 2025, and live authenticated agent transactions followed in Hong Kong (March 27, 2026) and Thailand (April 7, 2026). On June 10, 2026, it launched Agent Pay for Machines, aimed at software paying software, with more than 30 partners including Adyen, Checkout.com, Cloudflare, Coinbase, Ripple, the Solana Foundation, Stripe and Tempo. Mastercard says payments can be as small as fractions of a cent and can settle across cards, bank accounts and stablecoins.

The partner list is the more revealing detail. A card network signing up Coinbase, Ripple and the Solana Foundation shows Mastercard doesn’t expect cards alone to win machine payments. It wants to sit in the middle of whichever rail does. For context, Mastercard processed more than 175 billion transactions on its network last year, so even millions of agent payments would be a rounding error for now.

Stripe, Tempo and the Machine Payments Protocol

Tempo, the payments blockchain started by Stripe and crypto investor Paradigm, raised a $500 million Series A at a $5 billion valuation in October 2025. Its mainnet and the Machine Payments Protocol (MPP) went live on March 18, 2026, with more than 100 integrated services and design partners including Anthropic, DoorDash, OpenAI, Shopify, Visa and Mastercard. MPP bundles many tiny payments into one on-chain settlement, which keeps costs low. In its first weeks, MPP settled about $25,000 across roughly 115,000 transactions. A $5 billion valuation against $25,000 in early volume shows what investors are paying for: a position in a market they expect, not the one that exists today.

Google AP2 and the Protocol Race

Google launched the Agent Payments Protocol (AP2) on September 17, 2025, with more than 60 organizations, including American Express, Coinbase, Mastercard, PayPal and Revolut. AP2 uses signed digital “mandates” as proof of what you authorized, and it supports cards, stablecoins and real-time bank transfers. Google also built an x402 extension with Coinbase, the Ethereum Foundation and MetaMask.

Table 3: The Six Main AI Agent Payment Protocols

The Six Main AI Agent Payment Protocols

Note: Visa’s Intelligent Commerce Connect supports TAP, MPP, ACP and Google’s Universal Commerce Protocol through one integration.

Six protocols in 13 months is a lot for a market that still measures card-based agent payments in the hundreds. The more interesting trend is consolidation: Visa’s one-integration approach and Mastercard’s partner list suggest the networks plan to sit above all of these standards rather than pick a winner. For shoppers, that’s good news, since your agent shouldn’t need to care which protocol a store uses.

Consumer Trust: People Let AI Shop, Not Pay

US Consumers

Visa’s survey with Morning Consult, run January 29 to February 6, 2026 with 2,000 US adults, shows a steep drop-off as AI moves closer to your money.

Figure 5: How Comfortable Americans Are With AI at Each Shopping Step


The 20-point drop between comparing prices and completing a purchase is the trust gap in one picture. People are fine letting AI do the work they find boring, but not the step where a mistake costs money. The 60% who want to approve every spend aren’t rejecting agents; they’re asking for the same control a card’s “confirm purchase” button gives them today.

Nearly 40% said an AI agent or tool had led them to buy something they wouldn’t otherwise have considered, which shows agents already shape spending even without paying. On who they’d trust, 36% picked bank-backed AI, 35% payment network-enabled AI and 28% independent AI agents. A separate Visa Trust Index, fielded by The Harris Poll in May 2026, found that 72% of US adults have tried an AI assistant but only 23% would trust generative AI itself to execute a payment. Both surveys were commissioned by Visa, which benefits from a finding that people trust payment networks more than AI firms, so the ranking deserves some skepticism even if the overall caution rings true.

UK Consumers

UK shoppers are even more cautious. An ACI Worldwide survey of 2,080 UK adults, run by YouGov in June 2026, found that only 19% trust AI assistants to make everyday purchasing decisions, versus 55% who trust a human expert.

Figure 6: What UK Shoppers Trust an AI Agent to Do


The split is sharper than the US data. UK adults will trust an agent to find a deal or stick to a budget, but trust falls by about 30 points once the task involves acting in their interest, guarding their data or fixing a problem. The same survey found 60% would stop using an AI shopping agent after one mistake, 44% wouldn’t trust one regardless of how much it saved them, and 59% wouldn’t trust any organization to manage AI-powered shopping and payments. That last figure is the hardest for the industry: it isn’t a brand problem that a better logo can solve.

Spending limits are the most likely bridge. A smaller GoCardless survey of 300 UK consumers in February 2026 found 64% are open to AI managing recurring payments, mostly with conditions, and 45.3% want to approve every payment. Small, capped, repeat payments look like the first place UK shoppers will let agents in.

The Generation Gap

Age is the clearest dividing line. In Visa’s US survey, 48% of Gen Z trust payment network-enabled AI, compared with 20% of Baby Boomers. A gap that wide usually narrows over time as younger users become a bigger share of spending, which supports the long-term forecasts more than any current adoption number does.

Merchant Readiness for Agentic Payments

Merchants are preparing faster than shoppers are adopting. Checkout.com’s 2026 research found nine in ten merchants are actively preparing for agentic commerce and 42% are already testing it, yet only about 3% of transactions involve AI agents. 76% say payment providers will decide how fast agentic commerce can scale, and 25% of consumers would stop using an agent if a purchase was hard to dispute.

PayPal’s survey of 498 US merchants, run February 23 to March 3, 2026, highlights the liability gap. Nearly two-thirds said a standard liability framework, similar to today’s chargeback rules for online card payments, is urgently needed. About one in five merchants have 80% or more of their catalog available as structured, machine-readable data, which agents need to shop accurately.

Put together, the merchant data reads like a market preparing for demand it can’t yet see. Testing is cheap and the fear of missing a new sales channel is strong, which explains the 90% figure. But the catalog and liability numbers show the hard work is still ahead: an agent can’t buy what it can’t read, and a merchant won’t welcome agents until it knows who pays when one gets an order wrong.

Fraud, Bots and Liability

Agents create a new fraud problem: a good bot shopping for you and a bad bot testing stolen cards can look identical. The latest data shows why merchants are nervous:

  • DataDome tracked 52.7 billion AI agent and crawler requests between July 2025 and June 2026, up 82.3%. In H1 2026, 605.6 million AI bot requests hit high-risk pages such as logins, carts and payment forms.
  • The payment-flow share of AI traffic rose from 2% to 6.4%, more than tripling, and AI agent spoofing (bad bots pretending to be known agents) rose 45% between February and July 2026. DataDome says 80% of AI agents don’t properly identify themselves.
  • LexisNexis Risk Solutions saw agentic traffic rise 450% during 2025, mostly tied to card payments and logins on gaming and gambling sites, though it found no sign of malicious intent.
  • HUMAN Security measured 7,851% growth in agentic traffic in 2025 from a low base, with checkout pages making up 2.31% of agent activity.

The 80% figure is the one to watch. If four in five agents don’t say who they are, merchants can’t build rules that welcome good agents and block bad ones, which is exactly the gap Visa’s TAP and Mastercard’s credentialing tools are trying to close. Until those standards spread, the safest move for a store is to treat unknown agents like unknown bots, and that will slow legitimate agent shopping too.

On the crypto side, BeInCrypto counted two incidents that sent a combined $281,200 to attacker-controlled wallets after agents’ budgets were redirected. And when things go wrong, shoppers know who they’d blame: 54% of UK adults in the ACI survey said the AI company that built the agent should handle refunds, compared with 9% for the retailer and 3% for banks or card issuers. That puts AI firms on the hook for a role card networks have spent decades building. Banks are already pushing back on unverified access, and JPMorgan shut the door on AI agents like Muse logging in with customer passwords.

The US Pays for AI, China Runs It Free

Chainalysis’ geographic data shows who is behind agent payments when wallets can be located.

Table 4: US vs China Share of x402 Agent Payments, by Category

US vs China Share of x402 Agent Payments, by Category

Note: Chainalysis says most China-linked security payments went to AgentLISA, an AI smart contract auditing service.

The inference split is the headline, but the full table tells a more balanced story. Chinese developers aren’t absent from agent payments; they simply don’t pay for AI answers because they lean on free open-weight models, AI models anyone can download and run. They do pay for audits, market data, and payment tools, the services they can’t easily run themselves. That makes x402 data a good mirror of each country’s AI habits: the US rents its AI, China owns its AI and buys the extras.

Funding and Investment in Agent Payment Infrastructure

Investors are betting on the plumbing, not the bots. BeInCrypto found that 92% of tracked funding in this space went to payment infrastructure. That’s about $3.91 billion, compared to far less for trading agents. Tempo’s $500 million round at a $5 billion valuation is the largest single example.

That bet makes sense, given everything above. Right now, nobody knows which agent app will win. However, almost every agent will need to pay, prove what it was allowed to buy and settle in dollars. For public companies, the exposure is still indirect. As of October 2026, neither Visa nor Mastercard reports agent-related revenue. These figures show where money flows, not which stock or token will benefit. This is not financial advice.

FAQs

Do you need crypto to let an AI agent shop for you?

No. Agents that buy products for you, such as those built on Visa Intelligent Commerce or Mastercard Agent Pay, use the cards and wallets you already have. Crypto mostly runs in the background, when software pays other services a few cents for data or AI answers. Even if you never hold a stablecoin, you can still use an AI shopping agent.

How can you limit what an AI agent spends?

Most agent payment tools let you set a spending cap, require approval for each purchase or limit the agent to certain stores. Card networks also issue agent-specific tokens, digital stand-ins for your card number that can be switched off without canceling your real card. Start with low limits and approvals turned on until you’re comfortable with how the agent behaves.

What happens if an AI agent buys the wrong thing?

It depends on how the agent paid. A card purchase usually goes through your card’s normal dispute process, though the rules for purchases you authorized an agent to make are still being worked out. Stablecoin payments generally can’t be reversed once they settle, so a refund depends on the seller. Keep order confirmations and check your agent’s purchase history regularly.

How is an AI agent different from an AI chatbot?

A chatbot answers questions; an agent takes actions. Ask a chatbot for the cheapest flight and it lists options. Ask an agent and it can compare fares, fill in your details and book the seat, sometimes paying for it too. That ability to act, especially with your money, is why agents need extra safeguards such as approvals and spending limits.

Why can’t card networks handle tiny payments?

Every card payment carries fixed processing costs, plus fees shared between banks, card networks and payment processors. On a $50 purchase those costs barely register, but on a one-cent payment they can be many times larger than the payment itself. Stablecoin networks with fees of a fraction of a cent, and tools that bundle many small payments into one, fill that gap.

Are there laws that cover AI agent payments yet?

Not specifically. In the US, existing consumer protection and card rules apply to agent purchases, but they weren’t written with AI buyers in mind. The GENIUS Act, the federal stablecoin law, sets rules for issuers like Circle and is due to take effect no later than January 18, 2027. Meanwhile, card networks are writing their own standards for verifying agents and recording what users approve.

References

  1. ACI Worldwide. (2026, June 29). UK consumers’ trust in AI shopping agents [Press release]. Business Wire. https://investor.aciworldwide.com/node/26576
  2. Adobe. (2026, January 7). Adobe: Holiday shopping season drove a record $257.8 billion online with consumers embracing generative AI tools. Adobe Newsroom. https://news.adobe.com/news/2026/01/adobe-holiday-shopping-season
  3. Bain & Company. (2025, December 17). 2030 forecast: How agentic AI will reshape US retail. https://www.bain.com/insights/2030-forecast-how-agentic-ai-will-reshape-us-retail-snap-chart/
  4. Brave New Coin. (2026, October 15). Coinbase’s x402 records 138.5M transactions: Chainalysis shows AI agents reshaping crypto payments. https://bravenewcoin.com/insights/coinbases-x402-records-138-5m-transactions-chainalysis-shows-ai-agents-reshaping-crypto-payments
  5. Brohan, M. (2025, December 9). Morgan Stanley: AI agentic shoppers could drive $385 billion in online sales. Digital Commerce 360. https://www.digitalcommerce360.com/2025/12/09/morgan-stanley-ai-agentic-shoppers-385-billion-online-sales/
  6. Chainalysis Team. (2026, October 14). North America: Inside the growth of agentic commerce. Chainalysis. https://www.chainalysis.com/blog/north-america-crypto-adoption-2026/
  7. Checkout.com. (2026). Agentic commerce 2026. https://www.checkout.com/guides-and-reports/agentic-commerce-2026
  8. Crypto Briefing. (2026, October 15). x402 stablecoin payments see 152% jump in monthly active senders to 43,100. https://cryptobriefing.com/x402-stablecoin-payments-active-senders-surge/
  9. DataDome. (2026, October 7). State of bot and agent security report: Key AI findings. https://datadome.co/threat-research/state-of-bot-and-agent-security-report-key-ai-findings/
  10. First Page Sage. (2026, October 22). Agentic commerce statistics: October 2026. https://firstpagesage.com/seo-blog/agentic-commerce-statistics
  11. Forkast. (2026, September 28). The trust paradox: Agent payment infrastructure is outpacing consumer readiness. https://forkast.news/the-trust-paradox-agent-payment-infrastructure-is-outpacing-consumer-readiness/
  12. GoCardless. (2026). Agentic payment adoption. https://gocardless.com/guides/posts/agentic-payment-adoption
  13. Juniper Research. (2026, April 7). Agentic commerce set to generate $1.5 trillion globally by 2030. FinTech Futures. https://www.fintechfutures.com/press-releases/agentic-commerce-set-to-generate-1-5-trillion-globally-by-2030-as-payments-infrastructure-leaders-revealed
  14. Juniper Research. (2026, July). Paying at agentic speed: Inside Visa, Mastercard and Stripe’s agentic commerce push. https://www.juniperresearch.com/resources/blog/paying-at-agentic-speed-inside-visa-mastercard-and-stripe-s-agentic-commerce-push/
  15. LexisNexis Risk Solutions. (2026, March 26). Synthetic identities and agentic bots posing as human contribute to 8% global rise in fraud attacks. https://risk.lexisnexis.com/global/en/about-us/press-room/press-release/20260326-ccr-global-fraud
  16. Mastercard. (2026, July 30). Earnings review Q2 2026. https://www.mastercard.com/us/en/news-and-trends/stories/2026/earnings-review-q2-2026.html
  17. McKinsey & Company. (2025, October 17). The agentic commerce opportunity: How AI agents are ushering in a new era for consumers and merchants. https://www.mckinsey.com/industries/retail/our-insights/the-agentic-commerce-opportunity-how-ai-agents-are-ushering-in-a-new-era-for-consumers-and-merchants
  18. McGleenon, B. (2026, October 14). The uncomfortable truth behind 90% of AI payments. BeInCrypto. https://beincrypto.com/uncomfortable-truth-ai-agent-payments-2026/
  19. PayPal. (2026, April 27). What merchants really think about agentic commerce. https://www.paypal.com/us/brc/article/agentic-commerce-pulse-report-findings
  20. PPC Land. (2026, June 2). AI agents are now buying things – and fraud looks identical. https://ppc.land/ai-agents-are-now-buying-things-and-fraud-looks-identical/
  21. PYMNTS. (2026, September 15). TRM Labs analysis of x402 settlements. https://www.pymnts.com/?p=4193315
  22. Rijo, L. / Coin-Turk. (2026, September 4). Solana overtakes Base with 90% share in x402 transaction volume and count. https://en.coin-turk.com/solana-overtakes-base-with-90yuzde-share-in-x402-transaction-volume-and-count/
  23. Salesforce. (2026, January 8). AI and agents account for $262 billion of 2025 holiday spend. https://www.salesforce.com/news/stories/2025-holiday-shopping-data/

Vincee Cole is a technology journalist with four years of experience covering the full spectrum of modern tech — from consumer devices, artificial intelligence, to quantum computing, blockchain, and digital assets. His reporting cuts through complexity to deliver stories that are sharp, grounded, and relevant to both general readers and industry insiders. Previously, he worked with fintech research teams across Southeast Asia, analysing how emerging technologies are reshaping financial systems at scale.

AI Training Data Companies Overview in 2026: 10 Vendors Compared for Buyers

  • Nov 16
  • 16 mins read

Read more

AI Training Data Companies Overview in 2026: 10 Vendors Compared for Buyers

  • Nov 16
  • 16 mins read

AI Shopping Statistics 2026 Across Buyers and Brands

  • Nov 9
  • 23 mins read

Best AI Presentation Makers in 2026: 10 Tools Compared, Including the One You Might Already Own

  • Nov 9
  • 11 mins read

Best AI Summarizer in 2026: 8 Tools Matched to What You Actually Need to Read

  • Nov 9
  • 10 mins read

Best AI Tools for Finance in 2026: 9 Picks by Workflow

  • Nov 8
  • 8 mins read

Best AI Tools for HR in 2026: 9 Picks for Real HR Work

  • Nov 8
  • 8 mins read

Huawei Enjoy 100 Pro vs Pro Max: $30 More Buys 2,700mAh and a Sharper Screen, but Costs You the Better Selfie Camera

  • Nov 8
  • 11 mins read

10 Best Black Friday AI Gadget Deals in 2026: Top Picks

  • Nov 8
  • 7 mins read