Agentic commerce fraud: How to protect your online shop

This Insights series explores agentic commerce – a transformative shift where AI agents take the lead in shopping, payments and customer decision-making.

Following Amazon’s announcement of the “buy for me” agentic AI beta feature in April, as well as press releases by other companies, agentic e-commerce is already being touted as the harbinger of "a new era in e-commerce."

Many are excited about this development, which promises easier, better buying experiences for consumers and more opportunities for tech-savvy merchants to sell.

But what about fraud? Is a new era of agentic e-commerce fraud also around the corner? How will we be able to identify and stop bad AI agents while attracting and serving good AI agents?

Let’s talk about agentic e-commerce readiness for merchants, as well as what we can expect from agentic AI fraud and its prevention.

What is agentic commerce?

Agentic (e)commerce is an AI-powered way for consumers and companies to buy products. Setting a range of requirements, which can include price, specifications and other considerations, the buyer assigns intelligent AI agents with the task of tracking down the best and most suitable product and procuring it for them.

In most cases, the shopper completes the transaction and receives their items or services without even visiting the online shop in question. The agentic e-commerce market has been reported to have long-term potential of $1.7 trillion by 2030.

Convenience and speed are being advertised as the benefits of agentic e-commerce. Unlike consumers, the AI agents aren’t swayed by beautiful descriptions or attractive pages. Instead, they look at the features and price of an item to determine the best purchase for their user.

In agentic e-commerce, a human initiates the transaction and authorizes it – but they’re delegating the authority to actually execute it to an AI agent.

Coming up – soon

Note that at the time of writing, the infrastructure for agentic AI is being set up. Although there have already been several developments in the area, similar to the explosion of artificial intelligence and LLMs.

Therefore, our understanding of the process and details of agentic e-commerce might change in the near future – and with it, best practices for merchants and AI companies. They’re also likely to vary from agent to agent, and that’s where the concept of KYA – Know Your Agent – comes into the picture.

Agentic e-commerce is also expected to allow consumers to request refunds and returns, find and apply promotions and boost their shopping experience overall.

The implications of AI agents on fraud

For those tasked with safeguarding the security of payments, online accounts, consumers and companies from fraud and abuse, AI agents present some interesting complications.

With very few exceptions, up to today, bot-like behavior on an online shop or marketplace was indicative of suspicious criminal activity. This included brute-force account takeover scripts, card testing bots, and automated buying of event tickets for illegal reselling.

After the emergence of agentic e-commerce, automated agents visiting e-shops will now be very welcome, with merchants hoping to grab their attention through optimizing their descriptions and promotion.

To complicate matters even further, fraudsters have access to the same tools and will be able to create even more, similar agents that attempt fraud and abuse against companies. One version of this may be creating fraudulent agents to trick legitimate consumers, which would be the equivalent of a fake storefront. Another, simply misdirecting or trying to authorize otherwise legitimate agents.

In fact, 65% of fraud experts say they are already targeted by AI-enabled fraud, according to Ravelin’s latest research.

How, then, can we identify and stop bad AI agents, while ensuring we’re attractive to good AI agents?

Agentic e-commerce and false positives

If you – or your fraud prevention partner – don’t adapt your e-commerce fraud strategy, you are likely to see a substantial rise in blocked false positives.

This is because legacy anti-fraud solutions are likely to flag AI e-commerce agents as bots, considering this type of behavior was only linked to malicious activity in the past.

In reality, this will simply mean fewer sales, as more and more consumers will rely on a shopping method you are blocking.

What you can do today

In the age of AI-powered automation, AI-powered fraud detection is increasingly becoming a necessity. In addition to legitimate consumers using AI to purchase, fraudsters are also set to take advantage. Merchants predict that this will impact payment fraud the most, while the majority (66%) consider machine learning to be key in fighting fraud, according to Ravelin’s AI vs AI in Fraud report.

In terms of legitimate use by consumers, an e-commerce AI agent could be able to access a customer’s existing account with a retailer to complete the transaction. Or the agent may check out as a guest, using an API.

The good news is that despite initial appearances, a number of fraud signals remain the same, even when the shopping is delegated to an AI e-commerce agent and even when the agent is using the guest checkout option.

Much of the data that a sophisticated fraud solution will gather and calculate remains the same, and this means you can still get robust protection. The goal should be to accurately discern between trusted agents and non-trusted agents.

In fact, there are not as many unknowns as one might initially fear. Although the device and session data will differ from a human shopper, features that will help identify a trusted e-commerce agent include:

• Billing and delivery address
• Card velocity checks
• Location data
• Consortium data
• Payment method data
• Authentication attestation data
• Customer history (potentially)
• Network (potentially)

With the help of machine learning, you can calculate a fraud score and recommendation that takes all of the above into consideration, including the possibility the sale is conducted through agentic e-commerce.

Ensure you capture broad, deep data and information at several checkpoints of the customer journey. Only part of it will look different under agentic e-commerce. An estimated 80% of fraud signals will remain the same.

And if you use AI-powered fraud detection such as Ravelin’s, the regularly deployed new ML models will soon learn from the transactions they’ve seen, and identify risks associated with a rise in AI commerce agents with greater and greater precision.

For example, by collecting new device, user agent and browser identifiers from established good agents, these models will learn what looks like legitimate AI shopping agents and what looks suspicious.

Forward-looking vs. stagnating

Agentic e-commerce may represent a massive shift in how we buy online. But fraud professionals are already used to quickly adapting to change. We have the infrastructure to see data and discern the good from the bad.

Leveraging your data and AI-native solutions to calculate the level of trust that can be placed on a person or a transaction, merchants can continue to accept payments with confidence and thrive in the age of agentic e-commerce.

Finally, you might want to have a discussion with your fraud detection software vendor to ensure they are taking solid steps to both enable good AI agents to shop from you and strengthen your defenses from any new opportunities this brings for fraudsters.

This article originally appeared on Ravelin.com and is republished here with permission of the authors.

Previously in this series:

Part 2: AI agents are here – and they’re changing payments

How AI agents are rewriting the rules of shopping and decision-making.

Coming up next:

Part 4: Mapping the emerging models of agentic commerce

Explore how AI agents are redefining the shopper journey – from discovery to payment – across increasingly autonomous ecosystems.