Fighting AI with AI: Countering AI-Powered Fraud with the Dual Power of Experience and Intuition

Highlights: 

• Utilize supervised machine learning, trained on past fraudulent transactions and provider knowledge, to establish a baseline for identifying and flagging known fraud patterns, such as multiple purchases sent to similar masked addresses (malicious fraud/chargebacks).

• Apply unsupervised machine learning to detect emerging, unpredictable fraud by identifying "out-of-character" behavioral patterns that indicate high risk, which is critical for preventing account takeover, card testing, and potential data breaches.

While you may not be able to personally evaluate all transactions, AI can. It starts with the information you impart when customizing fraud detection and prevention for your company’s needs. A fraud prevention solutions provider, like Equifax, can further train detection systems with their knowledge. That experience creates a baseline for artificial intelligence to review incoming information and make near instantaneous decisions.

Supervised machine learning delivers further “experience” by looking over the history of failed or fraudulent transactions to determine risk going forward — on each and every attempt. This way, it learns from the myriad of attempts and gains experience that protects your business. This type of machine learning is all about studying the past to predict future risk.

In part one of this blog series, we looked at how AI is now being leveraged by sophisticated fraudsters to execute more automated attacks. In part two of this series, we examined how criminals use generative AI and automation to simplify complex schemes like promo abuse, card testing, and synthetic identity creation. And in this final part, we’ll detail how AI and machine learning tactics can be leveraged to fight AI-powered fraud attacks. 

Example: Malicious Fraud Resulting in Chargebacks

Criminals actively seeking to defraud your business rely on masking to attempt multiple transactions from many accounts. They then make purchases with stolen or compromised debit or credit cards, even digital payment options. After you ship the goods, the owner of the compromised payment method reports it as unauthorized with the issuing bank.

The card owner gets their money back. Your business takes the hit on lost revenue, logistics expenses, and time spent dealing with the situation. You even get chargeback fees. The fraudster gets your goods.

Since you can’t fight chargebacks from malicious fraud and get your money back, prevention is a critical component of protecting your bottom line. Elements that would stand out when reviewed by an experienced manager can tip AI off to a high-risk situation. Fraudsters might use AI to generate physical addresses that look similar, but have differences in capitalization, punctuation, or abbreviation. A postal employee could most likely get a package to the recipient using one of these generated addresses. Some label systems may even standardize and correct the addresses using non-standardized punctuation or capitalization prior to shipment. But we need to catch this type of situation before that point.

Fraud detection tools can standardize the addresses, recognizing additional risk due to attempts at masking. Identifying and refining these early on makes it much clearer when someone tries to make multiple purchases using different cards with goods going to the same address. AI-empowered systems then flag and review or decline the transactions, keeping your goods and revenue where they belong.

Using supervised machine learning, fraud detection systems learn from the history of previous purchases from your company and those protected by your solutions provider. This example is supervised because it gains experience by reviewing previous interactions where customers initiated a chargeback, but these systems can evaluate almost any situation with a known outcome. That’s part of what makes them so powerful in the fight against fraud.

Intuition: Using unsupervised machine learning to detect and prevent fraud

Intuition offers a solution for problems that experience can’t handle. In terms of modern AI, unsupervised machine learning provides that source of intuition. Computers may not have a gut to rely on when things “don’t look right,” but advanced machine learning checks for emerging patterns that may indicate fraud.

Unsupervised machine learning isn’t tracking which attempts complete, get declined, or result in chargebacks. It’s focused on underlying patterns and elements that go beyond using the past to predict future results. For example, you probably think there’s just innately something risky about a large number of transactions with the same or similar attributes. Intuition agrees.

Example: Account Takeover and Unauthorized Activity

Criminals engaging in fraud for profit often put the most effort into account takeover. By getting credentialed access to systems, they steal sensitive information, make purchases, alter account settings, and reveal stored debit or credit card numbers. Detecting compromised accounts remains one of the greatest challenges in fraud prevention. Experience, and supervised machine learning, only helps avoid security errors made in the past. It doesn’t predict potential future avenues of attack.

That’s where the intuition aspect of unsupervised learning comes into play. By looking for patterns of use that emerge from the data, it becomes clearer when a credentialed user acts “out of character” and suddenly tries to make large purchases of easily resold goods or changes to their accounts. Systems powered by AI recognize this and challenge or lock out the user for additional verification. Intervention can help prevent significant data breaches and unauthorized purchases.

Intuition aids similarly in preventing card testing. Even if each stolen identity uses different credentials, it would be very strange for so many different people to be tied back to the same locations, devices, or networks. Identifying this “strangeness” as risky activity helps shut down mass attacks quickly and effectively.

Equifax fraud protection solutions stand at the forefront of AI innovation

Everyone has their own, slightly different definition of what AI is. But regardless of differences in nomenclature, it should be clear to all of us that the fraudsters are coming at us faster than ever before. If we hope to stay ahead in this high-stakes game, we need comprehensive solutions that leverage:

• Best-in-class standardization and canonicalisation

• Event-level supervised machine learning: “Experience”

• Event-level unsupervised machine learning: “Intuition”

• Behavioral analysis for advanced risk management

• Ever-ready support and continuous improvement

The Equifax suite of fraud detection and prevention solutions combine these methodologies and deliver them to you through our product suite, so that you can remain competitive in the fight against fraud. Learn how our solutions seamlessly integrate with your existing processes to deliver industry-leading fraud prevention and protect your business.