Artificial intelligence learns to observe PIN codes entered at ATMs

A group of European researchersdeveloped a machine-learning method capable of determining from video the PIN code entered at an ATM. The system can crack the PIN even when the customer covers the keypad with their hand.

The method is based on a convolutional neural network and a long short-term memory (LSTM) module that processed video recordings of PIN-code entry on a test ATM. As a result, the system was able to guess a four-digit PIN with 41% accuracy and a five-digit one with three attempts at 30%. In testing, 58 volunteers participated.

During PIN-entry, the ATM display will not be hidden. Therefore they synchronised the appearance of asterisks on the screen with the movement of the hand for more accurate prediction of the digits.

During data collection to train the algorithm, the researchers asked right-handed volunteers to enter 100 randomly generated five-digit passwords covering all ten digits of the keypad. They ultimately collected 5800 video recordings of individual PIN-code entries.

There is currently no protection against this type of attack, according to the researchers. They argue that increasing the number of digits in a PIN makes it harder to remember the numbers, and a random arrangement of digits on the keypad would confuse ATM users. They add that their method works even when 75% of the PIN-entry panel is obscured.

In October criminals using a voice deepfake stole $35 million.

In September scientists managed to fool facial-recognition systems using natural makeup.

In August researchers found that AI is capable of composing phishing emails more effectively than humans.

Subscribe to ForkLog AI news on Telegram: ForkLog AI — all the news from the world of AI!