Where’s the stolen card? Generative AI can find it, says Mastercard. | PaymentsSource

Lionel Ng/Bloomberg

As more financial institutions consider how to best use generative AI, Visa and Mastercard are pushing the technology as a way to solve cybercrime and card theft schemes.   

Mastercard in late May updated its gen AI technology to speed the detection of compromised cards that are stolen through spyware, malware, card skimming or other methods. Earlier in the month, Visa added new gen AI tools that are also designed to combat card attacks. 

The card networks see an opportunity to expand their security services, which can aid in their diversification beyond fees from card swipes. Visa and Mastercard also have broader plans to expand the use of gen AI among clients for purposes such as customer service, developing digital products and producing marketing other content.  

“We are living in this rapidly expanding digital world. The beauty of this is it creates opportunities for unique experiences. And gen AI is in the middle of this,” said Johan Gerber, executive vice president of security and cyber innovation at Mastercard. “As with anything new there are always risks, and that is where we are looking.”

Mastercard is using gen AI to fill data gaps in identifying stolen cards. “The lines between fraud, cyber and financial crimes are blurring. The crooks at some point have to launder what they’ve stolen back into the banking system,” Gerber said. 

Mastercard partners with companies that obtain data from the dark web. To use stolen cards, crooks often leave only a partial trail on the dark web to cover their tracks. For example, some parts of an account number may be missing. Gen AI can analyze a trove of payments data to produce the potential missing card numbers faster and with more accuracy than was possible in the past, according to Gerber. 

“This way we can anticipate or find out which cards were compromised and tell the issuers that they need to take action on specific accounts,” Gerber said.  

Visa is also applying gen AI to security. A new AI-powered product is analyzing more than 15 billion yearly VisaNet payments to warn issuers when an account may have been compromised by an bot-driven attack. 

Generative AI can be used in several ways to improve payments fraud detection, according to Andras Cser, an analyst at Forrester. The technology can add synthetic transaction attributes — such as cumulative value, location and other data — to native transaction attributes to aid in risk assessments. 

“Gen AI also enables better and faster reporting by allowing investigators to form questions and get answers in natural languages,” Cser said. 

The security products from Visa and Mastercard come as the use of gen AI is still controversial with some banks and card issuers. Thirty percent of banks ban their employees from using gen AI, mostly out of concern about assessing its risk, according to research from Arizent, American Banker’s publisher.  

And the U.S. Department of Treasury in March issued a report on AI and cybersecurity that found banks are adopting gen AI and large language models slowly and for low-risk tasks such as code product and personal assistants such as pilots. 

At the same time, gen AI is gaining steam at finechs and payment technology companies, providing potential competition for banks. PayPal, Klarna, Amazon and Google are among the companies that are pairing expanding financial services businesses with gen AI to reduce staff overhead and to build services for consumer and business clients. Klarna, for example, has said the impact of gen AI has altered its hiring strategy. 

As gen AI is used to provide cybersecurity, and the technology that powers gen AI is improved, adoption should increase among banks, according to Gerber. “As AI progresses, I’m excited about what it will do for businesses,” he said. 

While the media includes lots of stories about how gen AI is being used to commit new types of fraud or to increase the scale at which fraud attacks can be conducted, they focus less attention on how gen AI is combating fraud, according to David Mattei, strategic advisor for the fraud and AML practice at Datos Insights. 

“And there are reasons to be optimistic with this technology,” Mattei said, adding that there is an improvement over traditional AI-powered fraud mitigation.

Most fraud-detection systems employ older forms of machine learning to identify fraud. These use a considerable amount of data to “train” the model, and the number of fraudulent payments is “sparse” compared to good payments, according to Mattei, who notes that traditional models work against a known set of bad payments, making it hard to spot new fraud attacks. “This creates the needle-in-the-haystack problem when building a model,” Mattei said. 

Gen AI can mitigate many of these issues, empowering fraud fighters to elevate their fraud-detection capabilities and create more flexibility in the models, according to Mattei. 

“Rather than relying on a data scientist to guide the learning process, gen AI is well equipped at discovering the unique characteristics of fraudulent transactions automatically and can do so using a larger quantity of training data than is feasible with traditional machine learning models,” he said.