An AI payment war is unfolding.

On June 16th, the AI version of Alipay was officially launched, named "Abao," offering tens of thousands of services that can be completed with a single sentence in a dialogue box. This marks the largest revamp in Alipay's history, with the new version already undergoing invite-only testing and set to gradually roll out to all users.

Just one day later, WeChat Pay officially released its AI exclusive card, enabling users to make consumption requests during conversations with Agents, handling everything from intelligent recommendations to order placement and payment.

Alipay and WeChat Pay have successively initiated an AI payment revolution, propelling domestic payment services into a new era of "conversation as transaction." However, both Alipay and WeChat Pay have emphasized the issue of fund management rights, which is of utmost concern to users.

Alipay stated that Abao assists users without touching their money, with AI only executing tasks explicitly approved by the user. Any changes in funds or payment processes require personal confirmation. WeChat Pay emphasized that the AI exclusive card is completely isolated from the main WeChat Pay account, with all consumption within the Agent limited to the exclusive card balance. Moreover, without the user's final authorization for each order, AI cannot spend a single penny.

The two payment giants are aggressively reshaping interaction and payment forms while also setting up stringent security firewalls with account isolation for AI. Finding a balance between efficiency and security, convenience and trust, may become the key test for AI payment to go mainstream.

I. The Arrival of the AI Payment War

During the golden age of mobile internet, Alipay and WeChat Pay became the two giants of mobile payments by leveraging their strengths in life services and social relationships, respectively. However, as the interaction revolution brought by large models reaches its midpoint, facing the potential threat of AI Agents possibly deconstructing all apps, both companies have almost simultaneously initiated an AI payment revolution.

Alipay's AI version, "Abao," is essentially an aggressive battle to defend its entry point. Over the past decade, Alipay's evolution has almost represented a typical form of China's mobile internet, with continuously superposition functions, expanding mini-programs, and extending services.

From a payment tool to a life service platform, and then to a city-level application entry point, as a super app with billions of users and millions of micro-ecosystems connected externally, Alipay, while providing all-inclusive convenience, has inevitably become bloated.

Those low-frequency but essential services hidden under third or fourth-level menus, used only once or twice a year, have gradually turned into isolated islands that are difficult to mine. Users not only have to remember paths but also frequently switch between different mini-programs and H5 pages.

The launch of the AI version, "Abao," is essentially a rewrite of the entire entry structure. Users no longer need to navigate through multiple menus or switch between dozens of tabs but can directly enter a dialogue box and complete tasks with a single natural language sentence.

From "checking housing provident fund" and "finding charging stations" to "helping me plan my travel route" and "where can I charge quickly nearby," capabilities originally scattered across different mini-programs, service providers, and pages are unified and folded into a single dialogue entry point.

Alipay hopes to fold its complex service system into a single dialogue entry point through a new interaction method, shifting from "people seeking services" to "speaking directly to reach services."

Compared to Alipay's aggressive revamp, WeChat Pay's strategy appears lighter.

WeChat's solution is to separately launch an AI exclusive card, allowing users to make consumption requests during conversations with Agents, handling everything from intelligent recommendations to order placement and payment. Currently, this card supports use in WorkBuddy, where users can generate a binding link by requesting to open a WeChat Pay AI exclusive card.

After binding, if a user makes a purchase request, the intelligent agent generates a plan. After confirming the order, an authorization page for the AI exclusive card appears, followed by the specific payment order. However, the final step of verifying and deducting the payment still requires manual confirmation by the user.

At the same time, UnionPay Business and JD.com have also entered the AI payment arena.

UnionPay Business's AI payment product allows users to complete transactions quickly by simply saying a natural language sentence or text instruction, achieving "payment by speaking" and "conversation as transaction." It is currently applied in three scenarios: park food reservation (park food reservation), cloud billing, and AI ordering.

JD.com released the Autonomous Agent Payment Protocol (A2P2), dividing autonomous agent payment into six levels from L0 to L5. L0 requires human confirmation for every payment, corresponding to traditional payment modes. L1-L2 belong to conditional automatic payments, which are the current mainstream AI-assisted payments. L3 supports agents autonomously completing small-value, high-frequency payments within a single task. L4 can handle complex tasks such as corporate procurement, household expenses, and multi-step business trips. L5 represents complete autonomous payment by agents.

JD.com aims to enable agents to autonomously complete payments under rule constraints.

Almost simultaneously, from the revamps of major companies' apps to the entry of national teams and e-commerce giants, the smoke of AI payment battles has already risen. The second half of mobile payments is accelerating from the era of scanning codes to the era of voice-activated payments driven by large models.

II. The Anxiety of Super Apps

Every aggressive turn by giants is often accompanied by deep-seated anxiety. Behind Alipay's all-out effort in full-end AI transformation and WeChat Pay's cautious entry with exclusive cards, both companies face their own limitations.

Alipay's biggest bet lies in the super entry mindset of "Abao." However, switching the existing user base of 1 billion users to a dialogue box with a single click is a relatively aggressive reform that is unprecedented in the industry.

Its core weakness lies in the persistence of its tool attributes.

First is the destiny of being a tool that is used and then discarded. Although "Abao" successfully folds nested mini-programs into a dialogue box, significantly improving the efficiency of complex functions such as checking medical insurance and withdrawing housing provident funds, users' deep-rooted perception of Alipay remains as a secure payment and financial management tool.

Second, low-frequency scenarios struggle to support high-frequency interactions. C-end users' demands for AI assistants are often high-frequency, instantaneous, and generalized, such as ordering coffee, chatting, retrieving information, and generating text. However, Alipay's core scenarios in healthcare, government affairs, and finance are inherently low-frequency but essential.

When users have generalized AI demands, their first instinct is often to turn to apps like Doubao or Qianwen. Breaking the mindset of "only opening Alipay when I need to pay or handle affairs" is the Inventory ceiling (existing ceiling) that "Abao" must confront.

WeChat Pay's choice to issue cards externally through "AI exclusive cards" and delegate interaction dominance to the WorkBuddy platform may be related to its product attributes.

Alipay has always been a financial tool serving transactions. Users open it to complete specific fund transfers. This money-driven attribute necessitates continuous proof of its irreplaceability as infrastructure, and AI is precisely the card it can use to break through growth ceilings.

WeChat, on the other hand, grew out of social relationship chains. Red envelopes and scanning codes are merely functions that conveniently emerged within social scenarios. Its moat has never been payment capability itself but the fact that a billion people are accustomed to chatting, sharing, and maintaining relationships here. This means that any adjustment to WeChat Pay's underlying architecture risks affecting not just payment scenarios but also the user experience of the entire social main scenario.

This is why WeChat Pay chose to test its AI capabilities within an independent WorkBuddy rather than revamping its entry point like Alipay.

Confining AI to an edge scenario has the advantage of almost zero trial-and-error cost. It won't shake users' overall trust in WeChat due to a single poor experience, leaving the social moat intact.

Wang Pengbo, a senior financial industry analyst, pointed out in an interview with Securities Times that WeChat Pay, born from the social ecosystem, has a far-reaching impact from underlying architecture adjustments and must prioritize stabilizing the main scenario experience. Therefore, launching a lightweight, physically isolated "AI exclusive card" is a Steady choice (prudent choice) to control trial-and-error costs and ensure fund security.

But the trade-off is that this path is destined to be slow. Currently, AI payment capabilities are only integrated into WorkBuddy, requiring binding to a third-party account for first-time use and only supporting Mac desktops (with more versions to follow), covering a very narrow range of users and scenarios.

From actual experience, compared to the direct order placement mode of traditional payments, users still need to go through multiple steps such as expressing demands, receiving agent recommendations, and authorizing confirmations. For users accustomed to scanning codes and one-click ordering, whether this new interaction method can truly improve efficiency remains to be validated across more scenarios and user habits.

III. How Far Is It from AI Spending Your Money?

Although the logic behind AI payments differs, none of the companies has fully handed over the final payment decision to AI.

Alipay stated that Abao assists users without touching their money, with AI only executing tasks explicitly approved by the user. Any changes in funds or payment processes require personal confirmation.

WeChat Pay emphasized that the AI exclusive card is completely isolated from the main WeChat Pay account, with all consumption within the Agent limited to the exclusive card balance. Moreover, without the user's final authorization for each order, AI cannot spend a single penny. WeChat Pay also revealed that this capability is undergoing small-scale testing in another product, QClaw, with plans to open it up to more Agent platforms in the future.

However, user reactions are not so optimistic. On social media, some complain that they "lack not payment methods but balance," while others straightforward words (bluntly say) that this interaction is too cumbersome and inferior to manual operation.

Behind such feedback lies the same question. Before AI can prove itself more convenient than manual operation, does adding a dialogue step truly save users trouble or create an additional barrier?

This precisely points to the same dilemma currently faced by both companies: finding a balance between security and usability.

If users are required to manually confirm every step, the process becomes cumbersome, undermining the significance of AI. However, simplifying the confirmation process means handing over more decision-making power to AI, which is inherently risky for a system that has yet to establish trust.

Both paths have currently chosen the former, minimizing risk but also leaving the hassle to users. This conservatism is not just a technical issue but an inevitable choice at this stage.

When agents execute payment actions on behalf of users, any misunderstanding can directly result in an incorrect deduction, which is a completely different magnitude of cost compared to AI answering a question incorrectly.

Will this conservatism persist long-term? The answer is likely no.

From the current product landscape, AI payment scenarios that can stand are still concentrated in low-frequency, lightweight areas with low trial-and-error costs, such as ordering food and making inquiries. Complex financial decisions are still not entrusted to AI.

However, this seems more like a starting strategy rather than the final goal. Platforms need to first accustom users to handing over decision-making power to AI in small-value, high-frequency scenarios, accumulating sufficient trust before potentially extending to deeper financial decision-making scenarios, such as proactively budgeting based on consumption habits, automatically comparing prices, and recommending optimal payment paths.

JD.com's recently released Autonomous Agent Payment Protocol (A2P2) provides a reference for observing this path. The protocol divides autonomous agent payment into six levels from L0 to L5, with L0 requiring human confirmation for every payment. As the levels progress, AI is allowed more autonomous decision-making space.

A relevant person in charge (person in charge) from JD Technology stated that this means agents are no longer just helping users add items to their shopping carts and waiting for manual payment but truly possess the capability to autonomously complete payments under rule constraints.

This level division to some extent confirms the previous judgment. The industry consensus may not be that AI can never autonomously spend money but rather that it's not time for that yet. Today's step-by-step confirmations are paving the way for tomorrow's autonomous agent payments.