Experian’s fraud chief has warned that highly sophisticated AI-created documents are rapidly reshaping mortgage-lending risk assessments.
Experian A/NZ head of fraud and identity Richard Atkinson has warned that AI-generated payslips, bank statements, and other income documents have become so persuasive that identifying fraud through manual review is now “nearing on impossible”.
His comments come after AUSTRAC’s Fintel Alliance revealed on Wednesday (19 August) that its Operation Claw project had uncovered co-ordinated suspected mortgage fraud after analysing data from 10 major Australian banks, identifying potentially hundreds of millions of dollars in suspect lending – largely linked to Sydney properties.
The financial crimes agency said that it had referred mortgage brokers, accountants, lawyers, and other entities to authorities as it expands its investigation into suspected mortgage fraud.
Atkinson said the emergence of increasingly capable generative-AI tools had materially changed the nature of documentary fraud, particularly across secured lending.
“What we’ve seen is the documents have become very sophisticated. We’ve seen a really tight correlation with when Banana Nano Pro came out, in November, that really stepped up the capability to create very accurate, very convincing documents,” he said.
Atkinson said that mortgage and vehicle finance were the two areas where Experian was seeing the greatest exposure and added that the quality of falsified material had advanced sharply in less than a year.
“It’s gone from even eight, nine months ago, the experts could read the documents manually by hand and could pick up some clues to these documents are now so convincing, if you’re looking at them just manually by hand, it’s nearing on impossible to detect the fraud,” he said.
“The other trend in that, not only is the tech making it near impossible to detect the fraud, the tech is in everyone’s hands, and it’s very simple for the man in the street to create that fraudulent document.”
Verification moves closer to source
Atkinson said brokerages increasingly had two broad choices: reduce their reliance on customer-provided documents by collecting information directly from their source or to use technology to identify manipulation within submitted documents.
“So what can you do in terms of particularly open banking, in terms of collecting the data directly from the customer’s bank account?” Atkinson said.
“If the customers used the open banking rails, given you access to their bank account, you know what their income is, at least their net income is. You can validate that. You can also validate their income and expenses.”
He noted that ATO data was not currently available through open banking, despite industry-wide calls for such information to become accessible through consent-based channels.
“The other path we’re seeing is to fight AI with AI. Now there’s cost to this, there are methods that very much uses AI, machine learning, image manipulation to take those documents and to read them and assess whether they have been manipulated,” he said.
“It is trying to stay one step ahead of those fraudsters. Those tools are very good at doing what they do, but you don’t detect every fraud.”
Original files matter
Atkinson also said that fraudsters could seek to remove evidence of digital editing by scanning or photographing a manipulated document before submitting it.
“Typical way fraudsters will hide behind that is you get a PDF of a statement, you manipulate it, then you scan it on your printer or on your phone,” Atkinson said.
“So what you’re providing is an image of a PDF. You’ve removed a lot of the information behind it. So from a broker’s perspective, you’re always requesting that the customer provides the original digital copy of that statement.
“That can be a great first move, and then provide those through to the lender, but these are tools that cost a lot of money.”
Atkinson noted that effective controls needed to assess whether an applicant was genuinely who they claimed to be, rather than relying only on the apparent presence of a driver licence.
“The banks are obviously dealing with a lot of applications; they are all under pressure to approve those applications faster and they want to do it in a cost-effective way,” he said.
“It always has to go back to, ‘Am I dealing with the individual that I think I’m dealing with?’ Moving beyond just citing a driving licence to performing complete checks – is that a valid driving licence? We’re taking a photo of the individual, are they the person presenting that driving licence?”
For direct channels, he said lenders could also assess digital signals such as IP addresses, VPN use, overseas access, repeat use of a physical address, and a device being associated with multiple applications.
Banks ‘running to catch up’
Atkinson said automation could prevent heightened fraud controls from creating unnecessary friction across every application.
“That works two ways. You automate that process, and all of those valid documents where there’s nothing wrong with it, brilliant automated process straight through. No one needs to look at, then those potentially invalid ones, yes, a human needs to look at it, and frankly, needs to go back to the customer and ask for more information to support the mortgage application,” he said.
“This is a challenge the banks weren’t prepared for, and now they’re running to catch up with where the situation is.”
Despite expected advances in digital identity verification, Atkinson said he expected fraud prevention to remain a continuing contest between lenders and criminals.
“Technology will take a step forward, but these fraudsters are taking steps forward as well,” Atkinson said.
[Related: AUSTRAC refers brokers to authorities in mortgage fraud probe]
Want to see more stories from trusted news sources?
Make The Adviser a preferred news source on Google.
Click here to add The Adviser as a preferred news source.
