Researchers trick ID scanners with fingerprints created by AI

American scientists have, through the application of artificial intelligence, created fingerprints that are relatively good at fingerprint scanners by tricking them into thinking that they are a real print.

New York University scientists have created artificially created fingerprints called DeepMasterPrints, which are able to imitate 23 percent of fingerprints in a biometric control system, the researchers report. That percentage becomes 77 percent if the margin of error used is increased from 0.1 percent to 1 percent.

DeepMasterPrints are able to do this because they specifically address two features of fingerprint-based authentication systems. First of all, not all fingerprint scanners scan the entire fingerprint, but only a part. This is the case, for example, with smartphones, where the scanner is quite small for ergonomic reasons and therefore cannot perform a full scan. Because small parts of a fingerprint are not as distinctive as a full fingerprint, a part of a fingerprint is more likely to be successfully matched with another stored fingerprint part.

In addition, there is the observation that, although complete fingerprints are unique, they also share certain properties with other fingerprints. This means that an artificial fingerprint with many of these common properties has a higher chance of being matched with other fingerprints.

Based on these two observations, the researchers deployed a neural network. Specifically, it concerns a generative adversarial network, which was trained with a dataset consisting of real fingerprints. The result was artificial fingerprints that not only have a relatively good chance of deceiving scanners, but also look real enough not to set off alarm bells if a person were to look at them. In a previous similar study, the artificial fingerprints still had spurious-looking corners, which would allow a human to filter them out.

According to the researchers, their underlying developed method is likely to find wide application in fingerprint security technology. The scientists argue that the artificial fingerprints could be used in the form of, among other things, a brute force attack, or specifically a dictionary attack, in which a large number of fingerprints are used to fake the verification of a control system.