Belgians warned of vulnerability to fake IDs from AI, digital copies

Belgium’s federal police have sounded the alarm about AI-driven identity fraud affecting individuals across the country.

The scams described in The Brussels Times are unsophisticated facial portrait swaps within photos of ID documents, which Belgian digital services and public transit ticket inspectors will accept as proof of identity.

Federal police told Flemish-language public broadcaster VRT there have been more than 10,000 victims within the past 12 months.

Authorities have warned the public that even a poor quality photograph of an ID card can be used to steal data, including the facial photo used in biometric matching by services like online loan applications and ecommerce sites.

VRT recounts the difficulties of an individual whose ID card was surreptitiously photographed two years ago. He has received dozens of fines for fare evasion on public transport since, as enforcement officers unquestioningly accept digital copies of his ID presented by perpetrators.

Transit operator SNCB-NMBS has begun introducing an ID verification software tool released by Belgium’s government called Checkdoc in response to the situation. In the meanwhile, the report from VRT shows a digital fake ID being accepted by some transit workers, and rejected by another in a part of the system where its frequent fraudulent use has been flagged.

The report notes the launch of the country’s actual digital ID later this year could improve the situation by providing a credential that can be digitally verified.

Protecting ID document-based verification

The frequent cases of unsophisticated ID fraud could be curbed by biometric verifications, but the exposure of ID documents can also fuel another, more sophisticated fraud method. Biometric facial morphing consists of combining two images so that the resulting image matches the biometrics of both individuals, and is one of the most pressing problems in biometric authentication today. Instead of local transit systems, this type of fraud can be used to defeat identity checks at international borders.

A pair of researchers from K. Ramakrishnan College of Engineering and the University College of Engineering in India have published their “design of a novel morphnet model for predicting image morphing and demorphing,” which will appear in the October issue of the journal “Biomedical Signal Processing and Control.”

The paper describes a method for “accurately reconstructing bona fide facial images from morphed inputs,” which in theory could allow investigators to identify and prosecute morphing fraud perpetrators.

The researchers developed a new dataset of partially manipulated faces with more than 17,000 images, and combined morphological filtering with convolutional feature extraction to improve performance in both morphing and demorphing. Their method reached 97 percent accuracy in Mated Morphed Presentation Match Rate (MMPMR) and showed resistance to existing morph detection techniques.

Article Topics

AI fraud  |  Belgium  |  biometrics  |  digital ID  |  face morphing  |  identity document