Bank fraud experts are using artificial intelligence (AI) to outsmart scammers who think they can fool financial institutions into believing they are customers and gain access to funds in accounts.

They are harnessing the technology to monitor scammers' behaviour and predict their next move.

With the assistance of machine learning and data analytics, banks here and the police have built systems that monitor transactions on an unprecedented scale.

The Singapore Police Force's Anti-Scam Centre (ASC) said it has integrated robotic process automation into its operations.

It creates a standardised format for the sharing of information on suspicious accounts with the banks, enabling them to forecast the next steps taken by bad actors.

Mr Royston Soon, vice-president of fraud risk management at OCBC Bank, said its systems are able to accurately identify transactions relating to scams.

"Using a machine-learning model, we are able to then rate the probability of the next transaction being fraudulent," he added.

"If the system detects a new behaviour on a new device, an alert will be triggered, and we can use this to identify money mules. We are then able to take action on such accounts even before the funds get transferred out."

Mr Soon said OCBC launched its fraud surveillance system in 2016.

The system monitors online transactions and gives each a risk score. High-risk ones trigger an alert, which is then reviewed by a fraud analyst.

He said that a machine-learning algorithm has enabled a higher probability of accurately detecting fraud cases.

Mr Soon added that OCBC also implemented its anti-financial malware system in 2019 which is able to identify what device its banking services are accessed from.

He said: "Even if you are using spoof technology, this system is able to identify your true IP and location. Fraudsters typically use one or a few devices to access multiple money mule accounts, which will then be flagged to our fraud analysts."

The system also captures behavioural biometrics, noting a user's typical finger movements and clicks on a device.

Mr Soon said the system would be able to capture one's mouse movements, typing speed and how a user navigates, backspaces and uses autofill components.

"We are able to then differentiate whether or not the one interacting on our platform is really our client," he noted.

Fraud analysts can then identify account takeover scenarios, and determine if it is a case of an account being used by a money mule.

He said the system is constantly learning as new data points are added.

Together with the information from the ASC, the bank has been able to create what is essentially a scam-hunting sentinel on its platform.

DBS Bank also has its own fraud detection system that uses machine learning and AI.

Mr Elvin Lim, head of group investigation at the bank, said: "We rely on machine learning and AI to basically identify transaction patterns of usage of customers' accounts, and using all the data points to tell us the story about whether this is a real customer or a potential scammer."

He pointed out that this network link analysis is like a spider web.

When one account gets hit, they analyse the kind of transactions that happen through that account.

The data is then processed and comes up in a pictorial format, helping to link other fraudulent accounts together and providing a comprehensive picture of entire scam operations.

Machine learning is also used to detect potentially fraudulent transactions at UOB, said Mr Richard Soh, head of investigations at the bank's Integrated Fraud Management department.

While these systems have been working well, they need constant updating as scammers change tack over time, he added.

"This is why we hope to strengthen our collaboration with the other banks through the ASC," he said. "A combined approach with the whole banking community is needed so that we can all harness and build on technology together to complement our efforts to stop crime."