Fingerprint authentication is one of the primary ways that most security conscious smartphone users opt to protect their devices against unauthorized access. This method gained mainstream popularity after Apple added Touch ID to its iOS-powered devices. The inclusion of Touch ID pushed Android device makers to widely adopt the authentication method as an essential feature. Now as the industry is gradually moving to facial identification as the primary authentication method, fingerprint authentication and fingerprint scanner have gradually gained a prominent space in the budget and entry-level Android-powered devices. However, a new report has indicated that Artificial Intelligence (AI) can be used to create something like a ‘master’ fingerprint to break the fingerprint authentication.
These ‘master’ fingerprints work similar to how ‘master’ keys or even backdoor passwords work that allow device or software makers to have access to any given device or service in case the user forgets their password/key. The cause of concern here is that here the master key is being used to beat a biometric sensor, something that has long been considered as the hallmark of security when it comes to authentication. This was initially reported by Motherboard where a group of researchers published a paper on arXiv outlining this research.
Watch: OnePlus 6T Thunder Purple Hands-On
The report did point that biometric authentication is not always secure but almost all previous attempts at beating it involved making a fake face, or fingerprint pattern that was somewhat matching to the owner of the device. This means that there was no ‘master’ key really to unlock multiple devices that were owned by different users.
According to the research paper, the fake ‘master’ fingerprint scanners “can serve as a match for a ‘large number’ of real fingerprints” that have been stored in databases. The paper termed these master keys as “DeepMasterPrints” where fingerprints from “over 6,000” people were fed to an AI neural network.
Researchers used a “generator” neural net to create these images and then used a “discriminator” neural net to separate fake fingerprints from genuine ones. In case the fingerprint was considered as fake then the “generator” made a small change and tried again for “thousands of times”. This continued till the “generator” was able to fool “discriminator”.
The researchers used two types of fingerprints, one taken from a capacitive fingerprint sensor and the other which used images of the fingerprints that were taken on a paper. Overall, prints from the capacitive sensor were significantly better at fooling the “discriminator” at all three levels of security.