US exchanges use AI to trade digital assets. Why is South Korea standing still?

The race to deploy AI agent payments is heating up in global financial markets. Payment companies and digital asset firms are moving to lock in infrastructure as AI takes over the comparison, trading, and direct payment of financial products on behalf of users. Meanwhile, analysts say South Korea's regulatory and market environment is falling relatively behind.

According to industry sources, global payment companies including Visa and Mastercard have moved beyond simply introducing AI agent payments and are now experimenting with accountability frameworks that allow agents to participate in transactions safely.

Visa struck a strategic partnership with OpenAI on June 10 to strengthen next-generation AI commerce, with the goal of enabling more secure payments in an agentic commerce environment. The two companies plan to connect Visa's global payment network and authentication and security infrastructure to OpenAI's services to build a foundation for AI agent payments. Visa will provide its payment network, tokenization, and risk management capabilities in the process.

Under the partnership, agentic payments will operate within permissions and policies set in advance by users — including spending limits, eligible merchant categories, and approval requirements. Actual transactions will use tokenized Visa payment credentials along with real-time approval and fraud detection systems. The aim is to preserve user control and payment security even when an AI agent is driving the transaction.

Visa had earlier launched the "Visa Agentic Ready" program to help card issuers prepare for AI agent payments, allowing them to test agent-driven transactions and assess their readiness for agentic commerce. In South Korea, six companies are participating: KB Kookmin Card, Samsung Card, Shinhan Card, Kakao Bank, Hana Card, and Hyundai Card.

Mastercard, which unveiled "Mastercard Agent Pay" last year to allow AI agents to make payments on behalf of consumers, this month introduced "Agent Pay for Machines." The service supports a range of payment methods from cards to stablecoins, with a focus on enabling AI agents to handle small, high-frequency transactions without human intervention.

Notably, Mastercard's partner list includes a large number of digital asset companies — among them OKX, the Solana Foundation, Ripple, MoonPay, Anchorage Digital, and BVNK. This signals that AI agent payments are expanding beyond traditional card networks to encompass blockchain-based payment infrastructure.

Behind this strategic expansion by global payment companies is a recognition that AI agents could upend existing payment accountability structures.

A survey conducted by Visa and PYMNTS Intelligence in January of 75 payment acquirers in Brazil, the UAE, and the United States found that 49 percent of respondents said resolving regulatory and compliance uncertainty was necessary to expand AI agent payments. Another 43 percent identified establishing standards that combine AI agent identity verification with payment credentials as a key challenge. The findings have led some market observers to conclude that global payment companies are racing to stake out positions in setting AI agent payment standards.

In the digital asset industry, Coinbase — the largest US exchange — is emerging as a leading example of connecting AI agent payment flows to stablecoins and on-chain finance. Coinbase is notably the company behind x402, a standard that enables payments without going through a card network. According to Coinbase, x402 has processed more than 185 million transactions over the past year.

Tiger Research said "the market x402 is targeting is different from Visa, Mastercard, and Stripe." Because card payments typically carry both a fixed per-transaction fee and a percentage-based fee, micropayments broken down to fractions of a cent — such as $0.01 or $0.005 — are difficult to process over card networks, the firm explained.

Coinbase is accelerating its push to build agent financial infrastructure. On June 17, it unveiled a corporate innovation strategy built around three pillars — trading, payments, and AI. On the AI finance side, it highlighted "Coinbase Advisor" and "Coinbase for Agents": the former provides investment advice based on a user's portfolio, while the latter connects a user's chosen AI agent to their Coinbase account to execute trades based on set conditions.

In South Korea, by contrast, the adoption of AI agent payments in the retail sector remains a distant prospect. What has been introduced so far is largely limited to natural-language financial task processing and internal workflow automation. There is still considerable reluctance to allow AI to exercise payment authority on behalf of financial consumers. Domestic digital asset exchanges are expanding AI-powered services, but these remain focused on information delivery and improving investment convenience rather than payments.

Park Hye-jin, a professor at Sogang University's AI and Digital Assets executive program, said that in AI payments, "what matters more than technical implementation is designing an operational structure that lets users use agents without suffering harm," adding that "refund processing, user authority delegation, spending limit settings, and building authentication systems are far harder challenges."

There are also calls for consumer protection frameworks to be established before AI agent payments can be introduced. Park said the financial authorities' stance is focused on financial consumer safety, adding: "Because the area where AI makes payments on behalf of people is closer to an operational and protection issue, it is inevitably burdensome to take the lead."

Hwang Seok-jin, a professor at Dongguk University's Graduate School of International Information Security, said South Korea "inevitably lags behind overseas in adoption speed due to a combination of regulatory and liability uncertainty, network separation requirements, authentication constraints, and other practical restrictions." Hwang added that "legal interpretation questions can arise over whether a payment made by AI on a user's behalf can be treated as the user's own expression of intent under civil law," and that "it is not clear who bears responsibility — the agent provider, the platform, or the payment company."

The gap in AI agent adoption between exchanges is also tied to differences in product variety, analysts say. Overseas exchanges are expanding the range of functions available to AI agents — including derivatives, tokenized assets, and prediction markets — beyond spot trading. South Korea, however, remains centered on spot trading, which limits the scope of tasks AI can execute. Hwang said: "In South Korea, there are no exchange-traded funds and only spot trading is available, so the payment system itself is inherently limited."

Data privacy and the cost of security investment are additional variables. A strengthened amendment to the Personal Information Protection Act is set to take effect in September, raising concerns that it could increase the cost of AI transformation for financial firms. Hwang said financial companies "need to pay attention not only to stablecoin and agent payment infrastructure but also to information security," stressing that "finding the right balance between technology adoption and consumer protection is crucial."

Starting in September, an amendment to the Personal Information Protection Act will take effect imposing fines of up to 10 percent of revenue on companies responsible for serious personal data breaches — tripling the previous penalty ceiling and extending liability to chief executives.