The fact that the fingerprint sensors on smartphones are not quite as secure as the manufacturers want us to believe, has already turned out to be the first iPhone with this feature. The Chaos Computer Club already made a copy of the matching fingerprint in 2013 using the simplest methods and was able to unlock it.
The technique has improved since then, the methods to crack it but also turn. The eternal cat and mouse game between team blue and team red. Now, a security team in the US has managed to create a kind of general key that allows 65% of all smartphones to unlock more or less easily. And not with the means of the analog, but the digital world – via machine learning and an artificial intelligence.
800 fingerprints as a template
The researchers used a data set of around 800 fingerprints, which showed a high degree of agreement with above-average other fingerprints. One can imagine this scheme as a key with a certain fit, which can be inserted into the corresponding cylinders, but can not open the lock.
They then produced a total of 8,200 partial prints from the record, which in turn were compared against each other. Say: all possible variants of whole fingerprints were created, which are possible with the combination of the partial imprints.
From this pool the researchers took the impressions, which corresponded with four percent of all other impressions in the database, so already could unlock one of 25 test-smart phones (statistically, the comparison lags a little, because not every human being in the world a smartphone with Fingerprint sensor).
These approximately 1,200 artificially generated impressions were in the next step the data basis for an artificial intelligence, which changed the templates arbitrarily minimally. If a change matched more pattern prints than the unchanged variant, the new version was retained. Quasi as if you are approaching the appropriate jigsaw puzzle with the fit and size of the jigsaw, or if you keep on a pattern key until you open the lock.
After a few rounds, the algorithm then created a general-fingerprint, a masterkey. This could unlock two out of three smartphones, although it was not the owner’s imprint. This functioned only in the best possible case, depending on the previously provided first impressions. However, it proves that the method can work without ever having to take the real fingerprint as a reference. For a potential attacker, the effort to get the original fingerprint in some way is no longer necessary.
According to the researcher, the problem with the scanner is that the built-in sensors are too small to check a whole impression at once. Rather, the correspondence of a partial imprint with a single further partial imprint would suffice to unlock the smartphone. On average, the devices store a total of approximately ten partial prints.
If you are interested in the details of the project, you can download the corresponding paper on IEEE.org. The team proposes to develop scanners that also scan grooves and differences in height of fingerprints. Some smartphones are already doing this – in the future hopefully further manufacturers will follow this advice ..