![[AI Magnifying Glass] AI selects items and makes payments… Can I leave my card with you?](/api/img?u=https%3A%2F%2Fimg.yna.co.kr%2Fetc%2Finner%2FKR%2F2026%2F10%2F02%2FAKR20261002022300017_02_i_P2.jpg&w=1200&q=72&f=webp)
Agentic commerce, in which AI agents go beyond recommending products and perform purchases and payments, is emerging, and domestic and foreign companies such as Naver and Mastercard are piloting related technologies, but consumers appear to prefer direct approval at the final payment stage.
AI-generated summary
AI agents were initially limited to product recommendations or price comparisons, but with recent advances in agentic commerce technology, attempts to automate even the purchase and payment process are increasing.
(Seoul = Yonhap News) Reporter Shim Jae-hoon = “Please catch a ride from Incheon Airport to Gwanghwamun.”
The artificial intelligence (AI) agent hears this and finds a means of transportation and reserves a vehicle. The user provides the destination and finally approves the payment, and AI handles the rest.
Until recently, the role of AI in online shopping was to recommend products that fit your tastes or compare prices.
However, as AI agent technology developed, the situation changed. Now, attempts are being made to entrust AI with finding products, putting them in shopping carts, and making payments.
The important point is that ‘choosing’ a product and ‘making me spend my money’ are different issues.
It is clear that it is convenient, but it is still unclear to what extent consumers will entrust their cards and accounts to AI.
◇ From “recommend” to “buy”… big tech competition
Recently, the interest in the global e-commerce industry is shifting to 'agent commerce', where AI goes beyond search and recommendations and even replaces purchases.
Existing conversational AI only answers users' questions and shows them products or purchase links, but in agentic commerce, AI identifies users' needs, compares multiple products, and is directly involved in the purchase process.
In September last year, OpenAI introduced 'Instant Checkout' based on the open standard 'Agentic Commerce Protocol' (ACP) developed with payment company Stripe.
However, last March, the strategy was revised, reducing stand-alone instant checkout where payment is made directly within ChatGPT, focusing on product exploration, and turning to allowing actual purchases to be made on a website or app operated by the seller.
‘Muse’, a personal AI agent released by Meta last month, can directly browse the website, find products, fill out forms, and process payment with the user’s approval right before purchase.
Looking at this trend, the focus of industry competition is not yet on whether to entrust the entire card to AI. The extent to which AI is given authority and at what stage it must be verified by humans is emerging as a key issue.
◇ Naver to pilot agent payment within the year… Even domestic actual transactions
Domestic companies are also speeding up.
Naver released a beta version of its shopping AI agent on the ‘Naver Plus Store’ last February. In June, a function was added to recommend products of interest first by analyzing user clicks and shopping cart records.
Within this year, we plan to pilot introduce a function to pay directly within the agent for some product groups.
The number of users also increased rapidly.
According to Naver, the number of monthly active users (MAU) and average daily conversations of the shopping AI agent last August increased approximately five times compared to March, when it was first launched. The average daily transaction amount through agents approximately quadrupled during the same period.
Digital and home appliances accounted for the largest transaction amount by item at 40%. This means that there was a great demand to utilize AI in fields where different specifications and functions of each product had to be compared.
There were also cases where AI actually paid money instead of people.
Mastercard announced that it completed the first domestic AI agent actual transaction using its payment solution 'Agent Pay' in March.
The AI agent searched for a means of transportation on the global mobility platform 'Hoppa', reserved a vehicle from Incheon Airport to a hotel in Gwanghwamun, Seoul, and completed the payment.
In the same month, Shinhan Cardo, along with Mastercard, also announced that it had demonstrated, for the first time in the domestic card industry, real transactions of 'AI Agent Pay' linked to global mobility services.
The AI searched for and reserved a means of transportation, and the user approved the payment once.
◇ “It’s okay to choose, but I have to pay”
The problem is consumer choice.
Consumers are relatively favorable to AI selecting candidates, but are negative to handing over the purchase decision itself to AI.
According to a survey of 322 U.S. consumers conducted by market research firm Gartner in January of this year, the response that they would entrust their purchasing decisions to AI did not exceed 11%, even for items such as household goods where the burden is relatively small if selected incorrectly. There were more responses saying that it was okay for AI to narrow down the options.
A similar trend was found in surveys of payment companies.
According to a World Pay survey of 8,000 consumers in the US, UK, France, Brazil, China, Singapore, and Australia, 80% of respondents interested in AI shopping responded that they could give AI some role in the payment process.
However, 59% said they would like to personally approve each purchase.
Even if you leave cumbersome tasks such as searching or price comparison to AI, you will check it yourself at the last moment when the money is spent.
◇ Who is responsible for incorrectly purchased items? Personal information protection is also a challenge
As technology advances, there are many institutional problems that need to be solved.
If the AI misunderstands the instructions and buys the wrong product or orders a product that is more expensive than the user wanted, it must be decided who will bear the damages.
If the seller's advertising costs or commissions affected the AI's recommendation results, it should be considered whether this can be considered a 'customized recommendation'.
When refund disputes or fraudulent payments occur due to hacking, it is not clear to what extent the AI developer, seller, payment agency (PG), and credit card company are each responsible.
Personal information protection is also an issue.
In order for an AI agent to select a product on behalf of a user, it must know purchase history and tastes, and in order to make an actual transaction, it must handle location, delivery address, and payment-related information.
An IT industry official said, “For agentic commerce to take hold, consumers must be able to clearly demonstrate when a dispute arises how much authority they have entrusted to AI.”
AI, which has been limited to shopping assistants, has now reached the payment button.
Technically, the era where AI uses money instead of people has already begun. However, we must keep in mind that it is up to us as consumers whether to hand over the authority to press that button to AI.
AI outlook — possibilities, not facts
Naver will pilot introduce an AI agent payment function for some product groups within the year.
Very likely · Within months
Consumers will likely prefer direct approval at the final payment stage for the time being.
Very likely · Within months

Antropic announced that it has revised its usage policy to block use of the generative AI Claude in cases of repeated and extremely offensive behavior. However, the specific criteria for judgment were not disclosed, and the debate over politeness and anthropomorphism regarding AI continues on social media.

Domestic security company Logpresso announced that it had confirmed that the Chinese suspect 'YY520CN', who was identified as the mastermind behind the financial sector hacking attack, was active as a community supervisor for the Chinese DDoS agency 'Godnet/Vitas'.

Coupang's shopping, food delivery, and streaming platforms experienced service disruptions around midnight due to an error during a service upgrade, affecting over 30 million monthly active users; access was restored by 2:30 a.m. after login and logistics issues persisted briefly.

Open AI's annualized sales were confirmed to be $50 billion, $20 billion less than the initially announced $70 billion, shocking the market, and this is believed to be due to differences in sales calculation methods.

President Trump will award the most prestigious medal in the field of science and technology to IT entrepreneurs including Musk, Jensen Huang, Lisa Su, and Sergey Brin at the White House. Fox News reported this as an example showing that the relationship between Musk and Trump has fully recovered.

CrowdStrike announced that it had discovered for the first time clues regarding the identity of the person believed to be behind a hacking attack targeting the Korean financial sector. It was estimated that the attacker may be a 26-year-old living in Guangdong Province, China, but it was determined that the identity of the attacker was not conclusive. The clue came from a request to write a resume for a security researcher exposed while using the generative AI coding tool 'Claude Code', and it was analyzed that personal information such as a Telegram account was entered.