Risk Control Shield | AI Shopping Surges Forward—Can Your Risk Control Hold the Line?

At the beginning of 2026, Alibaba’s internal trial of “zero-agent purchase” saw a daily API call volume surpassing 200 million—and this wasn’t part of a major sales event, but a regular day’s data.

During the Spring Festival, 130 million users experienced “one-click ordering” on the Tongyi Qianwen A「一」your command: ordering milk tea, takeout, or booking flights—from instruction to payment completion, the whole process took just 10 seconds.

Meanwhile, Taobao’s “Ask Me” feature can precisely recommend products after you describe a “matte lipstick for a date”; JD’s “Yanxi” can automatically compare prices across platforms when you’re hesitating; Meituan and Ctrip have fully integrated AI voice ordering…

Yet, in this process of AI reshaping the market, a neglected area is quietly heating up: fraudulent tactics are evolving in tandem.

AI shopping is reshaping the way consumers make purchases. Agentic shopping is moving from concept to reality. Consumers no longer need to go through the lengthy process of “browsing, comparing, adding to cart, and jumping to payment.” Instead, they can directly hand over their intentions to an AI assistant, which completes everything on their behalf.

From the user’s perspective, this represents the ultimate improvement in efficiency. But from the perspective of payment security, it brings a series of fundamental changes:

The paying entity is no longer unique
In the past, payment actions were performed by the cardholder themselves. Now, AI agents may initiate transactions on behalf of users. How do we distinguish between “AI agents authorized by the user” and “malicious programs impersonating the user”? This presents a brand-new identity verification challenge.

Transaction patterns are no longer stable
AI-driven shopping is often characterized by high frequency, small amounts, and high variability. The “user historical behavior baseline” that traditional risk control systems rely on is broken—because users themselves may not even know what their AI will buy.

The authorization chain has lengthened
In the chain of “user authorizes AI → AI executes the order → payment is completed,” a breach at any link can result in fraudulent transactions. Users might only discover that their AI “avatar” has been compromised when they receive their bill.

The data doesn’t lie—fraud is being transformed by AI.
The data from Black Friday 2025 has already sounded the alarm: shopping traffic generated by AI referrals surged 770% year-over-year. Over 68% of global consumers have begun using AI to make shopping decisions, or even authorized it to place orders on their behalf. At the same time, 69% of top North American retailers have already encountered AI-powered fraud attacks. Yet only 3% of retailers consider themselves “well-prepared.” What does this mean? It means that 99% of retailers may be defenseless against AI-driven fraud attacks.

In the AI era, fraud is no longer just about traditional tactics like stolen card or counterfeit card usage. It could be:

• Synthetic identities generated by AI, seamlessly bypassing traditional KYC rules;
• Prompt injection attacks that trick AI agents into executing unauthorized transactions;
• Mass-generated “authentic” accounts nurtured simultaneously on social and e-commerce platforms through AI;
• AI mimicking a user’s conversational style to deceive customer service or payment systems.

When attackers are armed with large language models, relying on traditional rule engines and manual reviews for risk control is like fighting missiles with bows and arrows.

The underlying logic of AI risk control is quietly shifting.
Traditional risk control revolves around rules + data, such as:
I have a blacklist and a whitelist, I set a maximum transaction amount, I enforce a rule like “no more than five transactions from the same IP within three minutes.”
Is that enough?
—Not anymore.

Because AI-driven fraud doesn’t follow predefined rules. Its attack patterns are dynamic, adaptive, and constantly shifting.
It may mimic the purchasing rhythm of real users, conduct small test transactions late at night, or instantly switch tactics the moment it detects being flagged.

In the AI era, fraud prevention is moving from rule-driven to behavior-driven —not about defining which actions are bad in advance, but determining in real time whether the current behavior belongs to the genuine user.

• Real-time behavioral modeling: focusing not just on historical data, but on behavioral characteristics every millisecond;
• Device and biometric fingerprinting: identifying whether the entity behind the screen is a real person or an AI program;
• Correlation network analysis: uncovering hidden connections between seemingly independent accounts;
• Self-evolving models: with every attack intercepted, the system learns and grows smarter.

The underlying logic of AI risk control is quietly shifting.
The year 2026 may well be remembered as the “first year of AI shopping.” Consumer habits are changing, transaction patterns are evolving, and fraud tactics are transforming.

The one constant is the balance between security and user experience.
As AI shopping becomes mainstream, risk control is no longer just a “gatekeeper” operating behind the scenes—it becomes the ticket determining whether a business can compete in this new era.
The wave of AI shopping is surging forward. Is your risk control system ready to ride it?