Onfido Fraud Lab Churns Out Fake IDs—and Trains Anti-Fraud Tech

Onfido has unveiled its “Fraud Lab”, in which a team of experts including former police and border agents have been working to train anti-fraud AI models over the past six months.

As Onfido explains, AI algorithms need to be trained on data from synthetic fraud attacks (such as those based on deepfakes) in order to learn how to defend against them, but the availability of such data can be limited. Onfido’s team has been dissecting and labelling the data from real attacks, and using generative AI to make synthetic derivatives, building a large-scale dataset that can effectively train AI models for optimal performance in detecting deepfakes attacks.

In other words, Onfido’s Fraud Lab team is picking apart real-world fraud attacks, making copies of them, and training anti-fraud models on all of that data. Onfido says this process allows its Fraud Lab to create as many as 10,000 synthetic IDs in just half an hour, enabling it to train and benchmark its anti-fraud AI on a large scale, and quickly.

Onfido has revealed these efforts alongside its publication of its 2024 Identity Fraud Report, which has found an alarming 31-fold increase in deepfakes and five-fold jump in digitally forged identities.

Unpacking the trends, Onfido Senior Fraud Specialist Simon Horswell said that a “gangster’s paradise” has emerged from the confluence of cheap and genuine ID data going up for sale on the dark web, and the rise of free and easy-to-use generative AI tools.

“Creating large-scale, convincing attacks at the click of a button has never been easier for malicious actors, who increasingly use AI to attack en masse,” he explained. “It’s become an AI vs AI battleground, where only defensive AI that’s trained on the latest attack vectors at scale can keep up with the evolving threat landscape.”

Further details from the 2024 Identity Fraud Report include a finding that digital document forgeries jumped 18 percent in 2023, and that about 80 percent of attacks aimed at spoofing a biometric liveness check involve the submission of a video of a video displayed on a device’s screen. The Report also find that document-only identity checks experience three times as many attacks as those that involve biometrics, suggesting that biometric tech remains a powerful deterrent for fraudsters.

That will come as welcome news to the many clients using Onfido’s own selfie-based identity verification and authentication solutions to guard their own businesses and data against fraud.

–

November 15, 2023 – by Alex Perala