Reviewed by Mayer Hyman, Payments Specialist | Reviewed for accuracy July 2026
Key Takeaways
- AI shopping agents that browse, decide, and check out on a customer’s behalf are moving from pilot programs to production. Visa expects millions of consumers to complete purchases through AI agents by the 2026 holiday season.
- Visa’s Trusted Agent Protocol and Mastercard’s Agent Pay (plus its new Agent Pay for Machines network) are the two major frameworks now defining how card networks recognize, verify, and authorize non-human buyers.
- Liability hasn’t shifted to the AI. Card networks and regulators are treating agent-initiated purchases as ordinary cardholder transactions, which means merchants can still eat the chargeback if an agent buys the wrong item, size, or quantity.
- Consent and price-match records are becoming the new fraud evidence: merchants who can prove the agent had explicit authorization and charged the price the customer saw are far better positioned in a dispute.
- Merchants don’t need to build agent support themselves. The practical first step is confirming that your payment gateway and checkout stack can recognize verified agent traffic instead of blocking or mishandling it as bot activity.
The Checkout Is Getting a New Kind of Customer
For the last two decades, “who’s buying” meant a person holding a card, a phone, or a login. In 2026, that assumption is breaking down. AI agents — the same class of tools now scheduling meetings and drafting emails — are increasingly the ones filling the cart and clicking “buy,” acting on instructions a human gave them minutes, hours, or days earlier.
This is agentic commerce: a human sets a goal (“book the cheapest flight to Chicago next Friday,” “reorder the printer ink when it’s under 20%”), and an AI agent executes the transaction without a human present at the moment of checkout. It’s a meaningfully different problem than the AI-powered product recommendations or chatbots merchants have used for years, because the agent isn’t just influencing a purchase — it’s authorizing one.
The card networks have concluded this is no longer speculative. Visa says AI agent-initiated purchases are moving “from experimental pilots to widespread consumer adoption,” and predicts millions of consumers will use AI agents to complete purchases by the 2026 holiday season. That is not a five-year forecast. It’s this year.
How the Card Networks Are Building Agent Infrastructure
Visa’s Trusted Agent Protocol
In October 2025, Visa launched the Trusted Agent Protocol with more than ten partners, including Cloudflare. The protocol’s core job is deceptively simple to state and hard to solve: give merchants a reliable way to tell the difference between a legitimate AI agent shopping on a customer’s behalf and a scraper or fraud bot hitting the same checkout endpoint. Before this kind of framework existed, most merchant security stacks treated all non-human traffic the same way — as something to block.
Mastercard’s Agent Pay and Agent Pay for Machines
Mastercard took a parallel path with Agent Pay, built on tokenization infrastructure similar to what already secures contactless and card-on-file payments, extended so a verified agent can transact using a consumer’s credentials without exposing the underlying card number. In June 2026, Mastercard went a step further with Agent Pay for Machines, a network purpose-built for machine-to-machine transactions — agents paying other agents or automated services, often in fractions of a cent, at a volume and speed no human checkout flow was ever designed to handle. More than 30 partners, including Stripe, Adyen, Coinbase, and Checkout.com, signed on as early supporters.
Both frameworks are converging on the same three requirements: verify the agent’s identity, confirm it has permission to spend on the consumer’s behalf, and set programmatic limits on what it’s allowed to spend without additional confirmation.
The Question Every Merchant Should Actually Be Asking: Who’s Liable When the Agent Gets It Wrong?
This is the part of agentic commerce that matters most for merchants, and it’s already being tested in court. In the Reyes v. Perplexity case, a February 2026 complaint documented roughly $1.8 million in Buy-with-Pro transactions that plaintiffs say were never actually reviewed or approved by the consumer before the agent executed them. Separately, the FTC filed a complaint against OpenAI in March 2026 alleging that Instant Checkout misrepresented refund eligibility and obscured merchant identity — a pattern Shopify linked to a reported 22% rise in disputed charges among affected merchants.
The regulatory answer, so far, is not “the AI is responsible.” The CFPB’s January 2026 advisory placed agent-initiated card purchases squarely inside the existing dispute-and-chargeback framework under Regulation Z, with one clarification that matters to merchants: a consumer’s right to dispute a charge isn’t eliminated just because they authorized an agent to shop for them — it’s only narrowed to the scope of what they actually authorized. If an agent buys the wrong size, the wrong quantity, or something outside the original instructions, that still looks like an unauthorized transaction to the card network, and the chargeback still lands on the merchant by default.
In practice, three things keep coming up as the deciding factors in these disputes:
- Was the listing accurate? If the agent bought based on inaccurate product information, that’s on the merchant.
- Did the agent have documented consent? A logged authorization showing the specific product, price, and quantity the consumer approved is quickly becoming the standard of proof — the agentic-commerce equivalent of a signed receipt.
- Did the price match? If the price the consumer’s agent was shown differs from what was actually charged, that discrepancy is a strong signal in the customer’s favor in any dispute.
None of this replaces existing fraud and chargeback tooling — it adds a new category of dispute for merchants and their payment partners to account for.
What This Means for Merchants Right Now
You don’t need an AI strategy to be affected by this shift — your checkout page is going to see agent traffic whether you’ve planned for it or not. A few practical steps matter more than chasing the trend:
1. Make sure your stack can tell agents from bots
If your fraud rules or bot-detection tools treat all automated traffic as malicious, you risk blocking legitimate agent-driven sales outright. Ask your payment gateway and security providers whether — and how — they support Visa’s Trusted Agent Protocol or comparable verification standards.
2. Tighten your listing accuracy
Agents check out based on what’s published: price, size, stock status, and description. Inconsistent or stale product data that a human shopper might catch and question is exactly what an agent will act on literally.
3. Treat consent records as part of your fraud evidence
Whatever documentation an agent platform provides about what the consumer actually authorized is likely to become standard evidence in future disputes. Merchants that can show the agent had explicit, scoped permission are in a materially stronger position than merchants who can’t.
4. Don’t overhaul your infrastructure preemptively
This is an infrastructure shift happening at the network level — Visa, Mastercard, and the major processors are building the plumbing. Most merchants won’t need custom integration work; the more urgent task is confirming your existing payment and fraud partners are keeping pace, not rebuilding checkout from scratch.
Frequently Asked Questions
What is agentic commerce?
Agentic commerce refers to purchases initiated and completed by an AI agent acting on a consumer’s behalf, based on instructions the person gave in advance, without the person present at the moment of checkout.
Is agentic commerce the same as a chatbot helping someone shop?
No. A shopping chatbot recommends products to a human who still clicks “buy.” An AI agent in agentic commerce actually completes the transaction — selecting the item, confirming the price, and authorizing payment — without a human clicking through checkout in real time.
Who is liable if an AI agent makes an incorrect purchase?
Under current card network rules and the CFPB’s January 2026 guidance, agent-initiated purchases fall under existing dispute and chargeback rules. Merchants remain liable by default unless they can show the listing was accurate, the agent had documented consent for that specific purchase, and the price charged matched what was disclosed.
Do merchants need to build new infrastructure to accept agent-driven purchases?
In most cases, no. Visa’s Trusted Agent Protocol and Mastercard’s Agent Pay are designed to work through existing card network and processor relationships. The practical step for merchants is confirming their payment gateway and fraud tools support agent verification rather than blocking it.
How big is agentic commerce expected to be?
Visa expects millions of consumers to use AI agents to complete purchases by the 2026 holiday season, and industry estimates put agentic commerce’s potential market size in the trillions of dollars globally by 2030, though actual adoption will depend heavily on how quickly trust, verification, and liability questions get resolved.
Cartis Payments helps merchants keep their payment infrastructure ready for what’s next, from fraud protection to gateway integrations that adapt as the checkout experience evolves. Talk to our team about preparing your payment stack for agentic commerce.
