AI-driven fraud hasn’t changed the need for humans, it’s changed what humans need to do

It didn’t take the fraud team at the major home improvement retailer to figure out something was up. 

Refund request after refund request, arriving complete with photographic evidence of a cracked plastic bidet seat. Same seat. Same damage. Same photo. Different customer credentials every time. And it turns out, a crack in the plastic that defied the laws of physics. 

Yes, AI was the culprit. AI being prompted by someone who was likely new at the game and didn’t realize even generative AI isn’t a magic fraud wand. 

Amid all the talk about the AI revolution, we’d be wise to remember the human is still in — or at least alongside — the loop. Humans at the home improvement retailer were key to exposing the scam. Humans within the fraud ring were responsible for the ham-handed, AI approach that blew their cover. 

None of which means that generative AI is not driving tremendous change in ecommerce fraud. In fact, fraud continues to grow faster than ecommerce sales by many measures. In 2026 compared to 2025, Signifyd data shows:

• Fraud pressure — the percentage of very risky online orders — was up 33%.
• Account takeover attempts increased by 78%
• Card testing spiked 175%

How is AI making fraud more lucrative? 

Meanwhile this year, Interpol reported that AI-aided fraud schemes were 4.5 times more profitable than non-AI-driven attacks and the FBI started breaking out AI-aided fraud attacks for the first time in its annual cybercrime report. 

While the increased malicious activity can’t be attributed entirely to AI, merchants who deny that AI is a contributing factor do so at their peril. 

My company’s State of Fraud Report 2026 detailed examples of how AI is industrializing fraud. Organized fraud rings once required teams of people with differentiated skills and time-consuming trial-and-error efforts to be successful. Now much smaller teams can match those efforts with significantly greater speed at a significantly greater scale. 

And so what it comes down to in 2026, is a discussion with a familiar ring — and a twist. The familiar ring: The key to success is finding the optimal balance between humans and machines. The twist: Today that imperative goes for fraud protection and fraud commission. 

What role do humans play in AI-driven fraud?

Signifyd’s recent report details a look-a-like website phishing scheme, a buy online, pick up (and return) in store money-laundering fraud scheme and cases like the home improvement retailer example in which returners use AI to gin up photos of damaged products.

Two of those — the fake website and the counterfeit damage photos — required AI to provide the scale and speed needed to make the fraud schemes profitable. But they also needed human intelligence and intuition to cook up scams that could result in getting something for nothing. 

The third — the BOPIS attack — relied heavily on humans to return products purchased with stolen credentials to the store, while persuading a store associate to credit the refund back to their personal Apple Wallets rather than to the stolen credit cards they used in the first place. 

And let’s not forget that phishing schemes, automated or not, rely on victims often experiencing very human emotions and reactions: doubt, worry, a sense of urgency, temporary and misplaced gratitude. 

On the other side of the fraud schemes above, the good guys’ side, humans played a key role in disrupting the crimes. 

The look-alike sites came to light when human agents saw patterns in complaints from consumers who weren’t getting what they paid for on the fake sites.The first warning in the damage photos case came from experienced fraud analysts whose antennae went up. And a risk team leader back at headquarters raised a red flag at seeing the frequency of the odd BOPIS refund pattern at the retailers’ brick-and-mortar stores. 

What’s clear from all this is that the changes we’re all living through go beyond a matter of degree and are in fact a transformation. AI’s influence on fraud and commerce means the number of split-second decisions merchants face is on the rise. So too are the  inputs they need to consider to make those decisions. 

How has the role of bots in ecommerce transactions changed?

Behavior that looked suspect two or three years ago, might be safe today. A customer who appears trustworthy based on patterns from 2022, might be anything but in 2026. The simple thinking that bots are bad is no longer true. AI and agentic commerce have changed the game.

That hasn’t eliminated the need for human intelligence. It’s just changed the tools humans need to build a system that effectively approves good transactions quickly and preserves lasting customer value while derailing transactions and activities that lead to fraud losses. Think about how that plays out in the world of fraud protection. 

First, it’s clear that it takes AI to beat AI. When fraud rings are launching AI-driven attacks, merchants need an AI-driven response. In fact, in 2025, a Merchant Risk Council report concluded that merchants were rapidly implementing AI and machine-learning-powered tools to detect and prevent fraud.

“Survey data also show that adoption of these kinds of tools and techniques is likely to continue apace,” the organization’s 2025 Global Ecommerce Payments & Fraud Report said. 

The conclusion was based on survey data that showed merchants who said they were using, or would soon be using, AI or machine learning tools reached as high as 78%. There was a general pattern in the responses: The more sophisticated the AI tool mentioned in the question, the higher the adoption and about-to-adopt rate. 

What does effective fraud protection look like in the agentic commerce era?

Second in the agentic commerce era, it’s equally clear that as AI becomes more embedded in the way businesses operate, the more precise the decisions rendered by those machines must become. Time is passing by AI systems that provide an educated — a very well-educated — guess or inference. Think: This is almost certainly fraud. This is almost certainly a good order. 

Today, merchants require a rock-solid decision that takes actions based on the previously provided instructions cooked up by intelligent humans who are experts in their fields. 

An inference tips the balance of human-plus-machine too far toward the human given the speed of commerce in 2026. A human can’t keep up with making hundreds or thousands of decisions that need to be made in order to provide customers with a great experience while protecting the business from wrong-doing.

The question for AI in commerce is fast evolving from, “What should the system do” to “What outcome did the system produce.”

That isn’t a world in which humans sit out while machines take over. That’s a world in which each partner operates at its highest and best level to produce the optimal outcome every time. 

And, by the way, it’s the world we live in today.