Agentic AI is reshaping how payments work, creating huge opportunities but forcing providers to adapt fast as new protocols, risks, and standards emerge.
The payments world is on the brink of another big shift. Following announcements by Cloudflare, Open AI and Visa, Agentic AI for payments—autonomous AI that can initiate, authorise, and complete financial transactions on behalf of users—has arrived.
This brings tremendous opportunity for payment providers, fintechs, banks and digital marketplaces to play for, but it’s also a challenge that could completely rewrite the rules of ecommerce and retail. With the rules not yet written, the wrong move could be costly.
A protocol problem
Right now, there’s no agreed standard for AI-driven payments. In September and October, three rival protocols launched almost back-to-back: Coinbase’s X402 for Anthropic’s Claude was followed within a week by OpenAI and Stripe’s Agentic Commerce Protocol (ACP) for ChatGPT. Then came Visa’s platform-agnostic Trusted Agent Protocol (TAP).
That means if your customer uses ChatGPT, payments may flow through Stripe; if they use Claude, it could be X402. It’s a familiar story for anyone who remembers previous format wars like VHS vs Betamax or HD DVD vs Blu-ray. These examples have shown us that technical superiority doesn’t guarantee adoption, because convenience and commercial fit can matter even more.
For payment providers and marketplaces, this fragmentation means much work needs to be done. Supporting multiple protocols will be essential just to make AI transactions possible, and developers are already asking how many protocols they’ll need to integrate. The lesson from Open Banking and PSD2 is clear: interoperability drives trust and scale. Beyond the US, Europe and Asia are likely to want their own standards as well, so coordination will be key; otherwise, global adoption will be patchy and inconsistent.
European Open Banking succeeded because regulators set the rules; however, in the US, privately developed protocols have been slower to scale—and we’re already seeing tensions emerge, such as JPMorgan’s proposal to charge for access to customer account data, which would be illegal under EU data-access rules. Whoever sets the standards first will shape the rails, and payments professionals need to engage early before the frameworks are finalised.
Is this the death of credit cards?
The most obvious area of the payments journey to be transformed by this is checkout.
Here, AI agents could finally give account-to-account (A2A) payments the boost they’ve been waiting for. Cards have dominated e-commerce so far because they’re easy for humans: the typical checkout involves only three simple steps – enter card number, confirm details, authenticate. By contrast, A2A has always been clunky.
AI changes that. It doesn’t care about long IBANs or multi-step authentication. Transactions that humans find tedious happen instantly. That makes A2A cheaper and simpler for everyone, opening the door to lower-cost payments at scale. Over time, cards could become just one of several ways to access credit.
But there’s a catch: consumer protection. Cards have built-in chargebacks and fraud safeguards, whereas A2A doesn’t. This gap becomes more significant when transactions shift from humans to autonomous agents. This isn’t a new problem for A2A, but AI amplifies it: if machines are initiating payments, consumers need confidence that they have meaningful recourse when something goes wrong. Robust protections aren’t just a “nice to have” – they’re essential if AI-driven A2A is ever to become a mainstream payment rail.
Managing the risk
Giving AI access to customer accounts isn’t risk-free. A misconfigured agent could trigger unauthorised payments or operational errors, and traditional fraud models – based on human behavioural biometrics like mouse movements and typing speed – will quickly become obsolete.
The challenge for payment providers then will be to build in safeguards and customer protections, whilst ensuring it’s as easy as possible for customers to interact with their systems. For example, providers could let customers segregate parts of their accounts that AI can access to keep control and visibility – through tokenised permissions, limited-scope APIs, or embedded KYC/KYAI (Know Your AI) frameworks. Spend limits could be managed through sub-balances and restrictions, and the AI’s transactions could be restricted and monitored.
Who’s liable?
Here’s the tricky part: if an AI agent makes a mistake, who is responsible? Current rules assume a human is in control, so when an autonomous system acts, accountability gets murky.
Self-driving cars is the prime example of this. If the car crashes, is it the manufacturer, the software developer, or the “driver” that is responsible? The payments industry faces the same challenge.
Rather than waiting for regulators to catch up, the industry should be calling for clear liability guidance now. Setting these expectations early helps ensure consumer trust and positions the payments sector as a responsible partner in shaping AI-ready financial infrastructure.
So what’s the next move for the payments industry?
The opportunity afforded by Agentic AI is clear, making payments faster, cheaper, and simpler, but only if the industry is prepared. So, the priorities are clear:
- Get involved in standards – join or form working groups to ensure interoperability and influence the new rails.
- Rethink fraud and risk – design detection models for non-human, autonomous behaviour.
- Define liability and protections – clarify accountability and safeguards for consumers and merchants before incidents occur.
- Experiment responsibly – launch controlled pilots or AI payment sandboxes to learn without exposing customer funds.
And if payments professionals don’t take the lead, tech giants will define the rails – potentially locking the industry into closed ecosystems.
With the shift toward Agentic AI, payments well underway, collaboration between payment providers, fintechs, regulators, and AI developers is now essential to ensure the next generation of financial infrastructure is open, safe, and customer-first.
