AI vs. AI: How APAC Banks Are Fighting New Fraud Types

The financial services industry is evolving fast. And so is fraud.

Take “friendly fraud.” A staggering 42% of Gen Z consumers now engage in it – deliberately disputing legitimate transactions they actually made.

It’s not a glitch in the system; it’s now part of a new financial landscape where AI-powered attacks and digitally-native dishonesty are becoming the norm across Asia Pacific’s banking sector.

“First-party fraud is the proper term,” explains Ian Holmes, global lead for enterprise fraud solutions at SAS Institute. “It’s your own customers acting against you.” This revelation is just one of many signaling a seismic shift in how financial fraud is evolving throughout APAC.

AI is not helping. Today, synthetic identities and deep fakes are becoming as common as counterfeit currency once was. And as banks deploy increasingly sophisticated AI systems to catch fraudsters, the criminals are using the same technologies to refine their attacks, creating an arms race where the prize is trillions in financial assets.

Synthetic identities, real problems

The digital era gave birth to a particularly insidious threat: the synthetic identity — a Frankenstein’s monster of stolen data points, fabricated information and AI-generated characteristics mixed with true identity data that can fool even sophisticated security systems.

“Synthetic identities are coming through the digital channel,” Holmes notes. “We utilize a lot of the device information in our machine learning to adapt to the nuances of how that device is being used.”

SAS’s approach focuses on what Holmes calls “proof of life” – algorithmic detection of human behavioral patterns through how devices are held, typing speeds, and even the natural errors people make when navigating interfaces. Even mistakes become authentication factors: “If you are typing on your phone keyboard as you go down the street, you’re going to make errors. Those errors are also proof of life.”

Beyond the head nod: Biometric integration

Taking a step further is the integration of biometrics with fraud detection systems, which has become increasingly sophisticated. Holmes explains that SAS typically receives biometric data from third parties, which is then incorporated into their machine learning algorithms.

“From a deep fake perspective, we’re able to harvest real-time data from the cameras within phones in order to do liveness checks,” Holmes explains. This includes countering “video injection attacks,” where pre-recorded footage might be used to spoof identity verification systems.

Standard liveness checks are frustrating, especially when you are asked to roll heads, blink and look sideways. To address this frustration, Holmes describes a new approach: “Rather than always asking the customer to do a certain activity, such as rotate their head, you can actually get them to do different things, like hold up two fingers... these cannot be built into the deep fake video itself."

Building the trust network

Perhaps the most significant evolution in fraud prevention is the development of cross-sector trust networks. These are collaborative ecosystems where banks, telecoms, government agencies, and other stakeholders share fraud intelligence.

“It's the new world. It really is the way that we have to go in order to be able to compete against the fraudsters,” Holmes insists. “The banks are seen to be the ones that are responsible for the losses of scams, but actually it’s weaknesses in the ecosystem that are resulting in the exposure.”

In Thailand, for example, mobile network operators have been mandated by regulators to ensure SIM cards are registered to the same person who holds the service contract, creating more reliable identity verification data that can be leveraged by financial institutions.

Federated learning: Divide and conquer

As data volumes expand exponentially, federated learning models are emerging as critical tools in the fraud prevention arsenal. Rather than building monolithic models across all data sources, federated approaches allow for specialized models focused on distinct data segments.

“If you build one model on all of that and a small piece of it changes, it’s disrupted all of your model from a governance perspective, whereas having federated models on each portion of that long message means that you can actually swap out that model very quickly,” Holmes explains.

The quantum frontline

On the horizon looms quantum computing and its implications for financial security. While SAS hasn’t directly invested in lattice-based cryptographic methods for transaction signing, Holmes notes they’re engaging with central banks on digital currency initiatives that leverage advanced cryptography.

“The ability to sign transactions digitally is something which is a core concept of the central bank digital currencies,” Holmes states, pointing to a future where quantum-resistant security measures may become standard across APAC’s financial infrastructure.

In such a future, having cleverer algorithms alone will not be enough. The future of fraud prevention will rely on how well we craft interlocking trust structures — bolstered by advanced AI and forward-looking cryptographic solutions — that can verify the increasingly tenuous concept of identity itself in a post-digital world.

Image credit: iStockphoto/wildpixel