TechUpdate

Credit card PINs can be guessed even while the ATM pad is covered

Researchers have demonstrated that it is possible to train a special-purpose deep-learning system to predict 4-digit card PINs 41% of the time, even while the victim is covering the pad with their hands.

The attack necessitates the establishment of a copy of the target ATM since training the algorithm for the exact size and key spacing of the various PIN pads is vital.

Using footage of individuals inputting PINs on the ATM pad, the machine-learning model is then taught to detect pad presses and assign specific probability on a set of possibilities. The researchers collected 5,800 recordings of 58 different people from various demographics entering 4-digit and 5-digit PINs for the experiment. The prediction model was run on a Xeon E5-2670 with 128 GB of RAM and three Tesla K20m with 5GB of RAM each.

The researchers reconstructed the right sequence for 5-digit PINs 30 percent of the time using three tries, which is generally the maximum allowed number of attempts before the card is withheld, and 41 percent of the time for 4-digit PINs.

The model may omit keys based on non-typing hand coverage and derive pushed digits from other hand motions by calculating the topological distance between two keys. The positioning of the camera that catches the attempts is critical, especially when filming left or right-handed people.

If the camera can capture audio as well, the model might employ pressing sound feedback that is slightly different for each digit, making the predictions much more accurate. This experiment demonstrates that covering the PIN pad with the other hand is insufficient to guard against deep learning-based assaults, but there are several remedies you may use. First, if your bank allows you to select a 5-digit PIN rather than a 4-digit PIN, go with the lengthier one.

Second, the proportion of hand covering considerably reduces prediction accuracy. A coverage percentage of 75% results in an accuracy of 0.55 for each trial, whereas entire coverage (100%) results in an accuracy of 0.33.

A third alternative would be to provide customers with a virtual and randomized keypad rather than the conventional mechanical one. This has unavoidable usability problems, but it is a great security precaution. Surprisingly, the researchers utilized video excerpts from the experiment in a poll with 78 people to see if humans could predict the hidden PINs and, if so, to what extent.

Sunaina

A tech enthusiast, with a mission to report data breaches, fraudulent practices, dark pattern practices, and updates. She is also frequently fascinated by fintech and unicorns.
Back to top button
Close
Close