Researchers from the Universities of Michigan and New York have created AI (artificial intelligence) fingerprints that are capable of unlocking most of today's smartphones. The researchers presented their work under the title: DeepMasterPrints: Generating MasterPrints for Dictionary Attacks via Latent Variable Evolution.
But how are Fingerprints created??
The fingerprints, which the researchers refer to as DeepMasterPrints, can be created using a machine learning algorithm. The fingerprints created by the algorithm were able to replicate more than 20% of the fingerprints in a database the researchers used, and they say the probability of error is one in a thousand replications.
The way the algorithm is able to create fingerprints that can fool smartphone sensors is due to two things. The first reason is that many sensors do not scan the entire fingerprint, but a small part of it. The second reason, which seems to be closely related to the first, is that in our fingerprints, there are some patterns that are quite common. So, patterns that are more common than others can fool the sensor.
The team trained a neural network to create virtual fingerprints, using innovative methods to find the most realistic DeepMasterPrint. The tests were conducted on 54,000 fingerprints from a public NIST.
However, such attacks cannot yet be carried out against smartphones. One of the researchers says that it would take a lot of work to reverse engineer such a system. However, they also said that hackers could certainly use this type of attack to gain access to applications and accounts. For this reason, they are trying to push organizations to adopt stricter security policies. In the case where it cannot be checked whether a biometric identity comes from a real person, then such attacks are possible.
